A coarse and fine wavelength division multiplexing combined multi-path parallel optical computing chip

Abstract

The application provides a multi-channel parallel optical computing chip combined with coarse and fine wavelength division multiplexing. The chip performs wavelength division multiplexing on multiple groups of input optical signals containing optical wavelengths through an optical computing input beam combining module, and guides the optical signals to a shared operation optical path. The optical signals are weighted and modulated through a microring resonator array in an optical multiplication and addition operation module, and the optical power is superimposed through a straight waveguide to complete the multiplication and addition operation. The optical signals carrying the operation results are demultiplexed through an optical computing beam splitting output module, and are guided to the corresponding output ports. The multiplexing and demultiplexing modules are realized by using the structural symmetry of the same waveguide Bragg grating array. Since the wavelengths in each group of optical signals respectively match the resonance wavelengths of the corresponding microring resonators under different free spectral ranges, multiple groups of optical signals can share the same group of microring resonator arrays. The chip effectively realizes n times parallel processing without additional hardware and power consumption.

Description

Technical Field

[0001] This invention belongs to the fields of optical technology and optical computing technology, and in particular to an optical convolution chip that combines coarse and fine wavelength division multiplexing. Background Technology

[0002] With the development of artificial intelligence and deep learning technologies, the demand for matrix multiplication and addition operations in neural networks is growing exponentially. However, traditional electrical processors are approaching their physical scaling limits, making it difficult to simultaneously achieve high computational output and low power consumption. Optical computing, with its inherent advantages such as ultra-high bandwidth, extremely low power consumption, and multi-channel parallelism, has become an excellent solution to break through the bottleneck of computing power.

[0003] Existing optical computing systems are mainly divided into coherent and incoherent architectures. Coherent computing networks based on Mach-Zehnder interferometers have large footprints, high insertion loss, and are prone to error accumulation. In contrast, incoherent computing systems based on wavelength division multiplexing have significant advantages in terms of space utilization and large-scale on-chip integration.

[0004] In the standard architecture of this type of system, the input data to be processed is first loaded onto multiple optical carrier signals of different wavelengths, and then these multi-wavelength optical signals are coupled into the same transmission waveguide through a wavelength division multiplexer. Next, these multiplexed optical signals are guided to a microring resonator array. Since each microring resonator has unique wavelength selectivity characteristics, it can accurately extract and respond to optical signals of specific wavelengths, thus enabling independent weighted modulation of each input signal. Finally, the multiple optical signals, after being weighted and modulated by different microrings, converge at the output, and the parallel multiply-accumulate operation result is directly output by superimposing the optical power.

[0005] However, wavelength division multiplexing systems based on microring resonators face severe physical limitations when scaling up computation. The parallelism of the system is fundamentally limited by the free spectral range of the microring resonator, thus preventing unlimited scaling.

[0006] The subject matter claimed in this disclosure is not limited to embodiments that address any shortcomings or embodiments that operate only in environments such as those described above. Rather, this background art is provided merely as an illustration of one or more exemplary technical fields in which some embodiments described in this disclosure can be practiced. Summary of the Invention

[0007] To address the problems existing in the background technology, this invention proposes a multi-parallel optical computing chip that combines coarse and fine wavelength division multiplexing. The optical convolution chip provided in this application can efficiently perform multiplication and addition operations between input data and convolution kernel weights by combining fine wavelength division multiplexing based on microring resonators with coarse wavelength division multiplexing based on waveguide Bragg gratings.

[0008] The technical solution adopted in this invention is: The present invention includes an optical computing input beam combining module, an optical multiply-accumulate operation module, and an optical computing beam splitting output module.

[0009] The optical computing input beam combining module receives multiple sets of input optical signals, wavelength division multiplexes each set of input optical signals, and guides them to a common computing optical path. The optical multiply-add module receives the optical signals in the computing optical path, performs weighted modulation and optical addition according to the convolution kernel weights, and outputs an optical signal carrying the convolution operation result. The optical computing beam splitting output module receives the optical signal carrying the convolution operation result, performs wavelength division multiplexing, and guides it to the corresponding output port. The multiple sets of input optical signals indicate the input data to be used in the convolution operation; the weighted modulated optical signal in the optical multiply-add module indicates the intermediate amount multiplied by the input data and the convolution kernel weights; and the optical signal carrying the convolution operation result indicates the final multiply-add result of the input data and the convolution kernel weights.

[0010] The optical computing input beam combining module and the optical computing beam splitting output module are the same waveguide Bragg grating array. The waveguide Bragg grating array contains multiple waveguide Bragg gratings connected in series through optical waveguides. Utilizing the structural symmetry of the waveguide Bragg grating array, it acts as the optical computing input beam combining module on the optical signal input side and as the optical computing beam splitting output module on the optical signal output side.

[0011] Specifically, the waveguide Bragg grating array includes a first input terminal group, a second input terminal group, a first output terminal group, and a second output terminal group. The multiple input optical signals are input to the waveguide Bragg grating array from the first input terminal group and to the optical multiply-add module from the first output terminal group. The signal output by the optical multiply-add module is input to the waveguide Bragg grating array from the second input terminal group and output from the second output terminal group.

[0012] The multiple sets of input optical signals are respectively input to each waveguide Bragg grating in the waveguide Bragg grating array through multiple input ports.

[0013] In this process, each set of input optical signals is reflected by a corresponding waveguide Bragg grating matched to its response band. The reflected optical signals are then transmitted along a series of optical waveguides and enter the through waveguide via the first waveguide Bragg grating. Each microring resonator modulates an optical signal of a specific wavelength. Optical signals not processed by the current microring resonator continue to transmit along the through waveguide. The modulated optical signals are then coupled into the coupling waveguides, where optical addition is performed to achieve optical convolution multiplication and addition operations before being transmitted to the last waveguide Bragg grating.

[0014] The optical multiply-add module includes a through waveguide, a coupled waveguide, and a microring resonator array.

[0015] The first waveguide Bragg grating in the waveguide Bragg grating array is connected to the through waveguide.

[0016] The last waveguide Bragg grating in the waveguide Bragg grating array is connected to the coupled waveguide.

[0017] The microring resonator array includes multiple microring resonators, each of which is disposed between the through waveguide and the coupled waveguide, and is optically coupled to both the through waveguide and the coupled waveguide, respectively.

[0018] Each of the aforementioned micro-ring resonators is equipped with a micro heater, which is used to thermally tune the corresponding micro-ring resonator to set the convolution kernel weight.

[0019] The present invention also includes multiple input ports and multiple output ports.

[0020] The multiple input ports are respectively connected to the input side of each waveguide Bragg grating, and are used to input the multiple sets of input optical signals.

[0021] The multiple output ports are respectively connected to the reflection output side of each waveguide Bragg grating, and are used to output the optical signal after wave decomposition and multiplexing.

[0022] The aforementioned multiple sets of input optical signals include n sets of optical signals, and each set of optical signals contains m optical wavelengths.

[0023] The microring resonator array comprises m microring resonators with a preset uniform spacing and resonant wavelength.

[0024] The m optical wavelengths in each group of optical signals correspond to the resonant wavelengths of the microring resonator array in different free spectral ranges, so that the n groups of optical signals share the same group of microring resonator arrays for weighted modulation.

[0025] The center wavelength spacing between adjacent waveguide Bragg gratings in the waveguide Bragg grating array is consistent with the free spectral range of the microring resonator. From the first to the nth waveguide Bragg grating, the center wavelength increases by one free spectral range with each additional waveguide Bragg grating.

[0026] The optical bandwidth of each waveguide Bragg grating is , where k is the preset uniform spacing of the resonant wavelength. The passband of the i-th waveguide Bragg grating covers the wavelength range of the i-th group of optical signals, matching the response of the corresponding microring resonator across the entire free spectrum.

[0027] The beneficial effects of this invention are as follows: 1. The optical computing input beam combiner and optical computing beam splitter output module are cleverly designed to share input and output ports using symmetry, thus utilizing the same waveguide Bragg grating array. The waveguide Bragg grating array, with its structural symmetry, acts as the output module on the optical signal input side.

[0028] This shared approach facilitates consistency in channel wavelength multiplexing and demultiplexing during system implementation, resolving inconsistencies in multiplexed and demultiplexed wavelengths caused by processing errors in actual production, thus improving computational accuracy. Simultaneously, it utilizes fewer components, reducing chip area and resulting in higher system integration.

[0029] 2. The center wavelength spacing between adjacent waveguide Bragg gratings in the waveguide Bragg grating array is consistent with the free spectral range of the microring resonator, and the center wavelength increases by one free spectral range in sequence; the optical bandwidth of each waveguide Bragg grating is "k×m", where k is a preset uniform spacing of the resonant wavelength, so that the passband of the i-th waveguide Bragg grating covers the wavelength range of the i-th group of optical signals, so as to match the response of the entire free spectral range covered by the corresponding microring resonator.

[0030] By coordinating the design of the optical computing input beam combining module, the optical computing beam splitting output module, and the optical multiply-accumulate operation module, the m optical wavelengths in each group of optical signals correspond to the resonant wavelengths of the microring resonator array under different free spectral ranges, so that the n groups of optical signals can share the same group of microring resonator arrays for weighted modulation.

[0031] This not only significantly reduces the number of microring resonators required, but also allows for greater system integration by using fewer components and reducing chip area.

[0032] Meanwhile, since the weighted modulation of the microring resonator array consumes power, and n groups of optical signals share the same microring resonator array, the computational load and computational speed are increased by n times without increasing energy consumption, which greatly improves the computational speed and computing power, while reducing losses and increasing integration.

[0033] 3. Significantly improves computing speed and the number of parallel channels.

[0034] By rationally designing the traditional method of optical computing based on microring resonators, the number of computing channels can be increased from 1 to n, increasing computing power and computing speed by n times without increasing power consumption.

[0035] Furthermore, the computing power of traditional optical computing based on microring resonators is limited by the total number of wavelengths, which is confined to a single free spectral range (e.g., only m wavelengths can be used, allowing for only m calculations at a time). This new design cleverly utilizes multiple wavelengths from different free spectral ranges within the microring resonator array of the optical multiply-accumulate module simultaneously through an optical input combining module and an optical output splitting module. With the same number of microring resonators, this allows for m×n calculations at a time, significantly increasing computing power and the number of parallel computing channels. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the structure of a multi-channel parallel optical computing chip that combines coarse and fine wavelength division multiplexing.

[0037] Figure 2 This is a schematic diagram of the optical path transmission of the optical computing input beam combining module.

[0038] Figure 3 This is a schematic diagram of the optical multiplication and addition module.

[0039] Figure 4 This is a schematic diagram of the optical path transmission of the optical computing beam splitter output module.

[0040] Figure 5 This is a schematic diagram illustrating the matching principle between the passband of a waveguide Bragg grating and the resonant wavelength of a microring resonator. Detailed Implementation

[0041] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0042] This application of the present invention provides a multi-path parallel optical computing chip combining coarse and fine wavelength division multiplexing. This chip uses an optical computing input beam combining module to combine multiplexed optical components... Multiple input optical signals of different wavelengths are wavelength divided and multiplexed, and then guided to a shared computing optical path.

[0043] This invention uses a micro-ring resonator array in the optical multiply-add module to perform weighted modulation of the optical signal, and combines this with a straight waveguide to perform optical power superposition to complete the multiply-add operation. The optical signal carrying the operation result is demultiplexed by the optical computing beam splitter output module and guided to the corresponding output port.

[0044] The multiplexing and demultiplexing modules are implemented using the structural symmetry of the same waveguide Bragg grating array.

[0045] Because the wavelengths of each optical signal are matched to the resonant wavelengths of corresponding microring resonators in different free spectral ranges, multiple optical signals can share the same microring resonator array. This chip effectively achieves n times the parallel processing without additional hardware or power consumption. Furthermore, the combination of fine wavelength division multiplexing based on microring resonators and coarse wavelength division multiplexing based on waveguide Bragg gratings overcomes the physical limitations of the free spectral range, significantly improving the scalability and computational density of the optical convolution chip.

[0046] The function to be achieved by this invention is to perform highly parallel photonic convolution multiplication and addition operations.

[0047] Since this application uses waveguide Bragg gratings to achieve coarse wavelength division multiplexing, enabling multiple sets of input optical signals to share the same microring resonator array, the core physical process of this operation can be abstracted as the parallel dot product summation of multiple sets of input vectors with fixed convolution kernel weights. Its specific matrix operation formula can be expressed as:

[0048] Where matrix X represents the input data to be convolved, which is loaded onto the optical signal.

[0049] This embodiment contains n sets of optical signals, and each set of optical signals contains m wavelength channels. Each wavelength constitutes Matrix X; convolution kernel weight vector W, representing the convolution kernel set by the micro-ring resonator array.

[0050] Since the n groups of optical signals in this invention share the same array containing m microring resonators, the transmittance set by the micro heater constitutes... The convolution kernel weight vector W; the vector Y represents the result of the multiplication and addition operation between matrix X and the convolution kernel weight W.

[0051] The final system outputs n independent optical signal results in parallel through the optical computing beam splitting output module, i.e., one Column vectors.

[0052] like Figure 1 As shown, the specific implementation of the present invention includes an optical computing input beam combining module, an optical multiply-accumulate operation module, and an optical computing beam splitting output module connected in sequence.

[0053] The optical computing input beam combining module receives multiple sets of input optical signals, wavelength divides and multiplexes each set of input optical signals, and guides them to a common computing optical path. The optical multiply-add module receives the optical signals in the computing optical path, performs weighted modulation and optical addition according to the convolution kernel weights, and outputs an optical signal carrying the convolution operation result. The optical computing beam splitting output module receives the optical signal carrying the convolution operation result, performs wavelength division multiplexing, and guides it to the corresponding output port.

[0054] Among them, multiple sets of input optical signals are used to indicate the input data to be convolutionally processed; the weighted and modulated optical signal in the optical multiply-add module is used to indicate the intermediate amount of the input data multiplied by the convolution kernel weights; and the optical signal carrying the convolution operation result is used to indicate the final multiply-add result of the input data and the convolution kernel weights.

[0055] Furthermore, the optical computing input beam combining module and the optical computing beam splitting output module reuse the same waveguide Bragg grating array.

[0056] The optical path transmission diagram of the optical computing input beam combining module is shown below. Figure 2 As shown.

[0057] This module is used to receive the above-mentioned n sets of input optical signals containing the data to be convolved, where the input optical signals correspond to the n rows of data in matrix X.

[0058] This module contains multiple waveguide Bragg gratings connected in series via optical waveguides. Each waveguide Bragg grating has a specific reflection passband.

[0059] Specifically, n sets of input optical signals are input into the waveguide Bragg grating array through n input ports respectively. Each set of input optical signals is reflected by the corresponding waveguide Bragg grating that matches its frequency band response. The reflected optical signals are transmitted forward along the series-connected optical waveguides and finally pass through the first waveguide Bragg grating, and are guided along the waveguide to the common computing optical path.

[0060] The structural diagram of the optical multiplication and addition module is shown below. Figure 3 As shown, it is responsible for performing weighted summation operations in the mathematical model, including through waveguides, coupled waveguides, and microring resonator arrays.

[0061] The first waveguide Bragg grating in the waveguide Bragg grating array is connected to the through waveguide, and the last waveguide Bragg grating in the waveguide Bragg grating array is connected to the coupling waveguide.

[0062] The microring resonator array includes multiple microring resonators, each of which is disposed between the through waveguide and the coupled waveguide, and is optically coupled to both the through waveguide and the coupled waveguide, respectively.

[0063] Each of the aforementioned micro-ring resonators is equipped with a micro heater, which is used to thermally tune the corresponding micro-ring resonator to set the convolution kernel weight.

[0064] Specifically, the optical multiply-add operation module micro-ring array contains m micro-ring resonators, which physically correspond directly to the m elements of the convolution kernel weight vector W.

[0065] Each microring resonator is equipped with a micro heater. When multiple optical signals entering the through waveguide flow through the microring array, the refractive index of the microring waveguide is changed by adjusting the thermo-optical effect through the thermo-optical voltage applied to each micro heater, thereby setting the transmittance of each wavelength channel. The modulated optical signals are then coupled into the coupling waveguides below.

[0066] The coupled waveguide directly superimposes the optical power of all modulated optical signals in space, completing the summation operation in the mathematical formula and generating an optical signal carrying the result of the convolution operation.

[0067] The optical computing input beam combining module and the optical computing beam splitting output module are the same waveguide Bragg grating array.

[0068] Therefore, the optical computing input beam combining module and the optical computing beam splitting output module are cleverly designed to share resources by using different input and output ports based on symmetry, thus utilizing the same waveguide Bragg grating array. The waveguide Bragg grating array, with its structural symmetry, acts as the output module on the optical signal input side.

[0069] This shared approach facilitates consistency in channel wavelength multiplexing and demultiplexing during system implementation, resolving inconsistencies in multiplexed and demultiplexed wavelengths caused by processing errors in actual production, thus improving computational accuracy. Simultaneously, it utilizes fewer components, reducing chip area and resulting in higher system integration.

[0070] The waveguide Bragg grating array of the present invention comprises multiple waveguide Bragg gratings connected in series via optical waveguides. Utilizing the structural symmetry of the waveguide Bragg grating array, it serves as the optical computing input beam combining module on the optical signal input side and as the optical computing beam splitting output module on the optical signal output side.

[0071] Specifically, the waveguide Bragg grating array includes a first input terminal group, a second input terminal group, a first output terminal group, and a second output terminal group. The multiple input optical signals are input to the waveguide Bragg grating array from the first input terminal group and to the optical multiply-add module from the first output terminal group. The signal output by the optical multiply-add module is input to the waveguide Bragg grating array from the second input terminal group and output from the second output terminal group.

[0072] The multiple sets of input optical signals are respectively input to each waveguide Bragg grating in the waveguide Bragg grating array through multiple input ports.

[0073] In this process, each set of input optical signals is reflected by a corresponding waveguide Bragg grating matched to its response band. The reflected optical signals are then transmitted along a series of optical waveguides and enter the through waveguide via the first waveguide Bragg grating. Each microring resonator modulates an optical signal of a specific wavelength. Optical signals not processed by the current microring resonator continue to transmit along the through waveguide. The modulated optical signals are then coupled into the coupling waveguides, where optical addition is performed to achieve optical convolution multiplication and addition operations before being transmitted to the last waveguide Bragg grating.

[0074] The aforementioned multiple sets of input optical signals include n sets of optical signals, and each set of optical signals contains m optical wavelengths.

[0075] The microring resonator array comprises m microring resonators with preset uniform spacing and resonant wavelengths. The m optical wavelengths in each group of optical signals correspond to the resonant wavelengths of the microring resonator array in different free spectral ranges, so that the n groups of optical signals share the same group of microring resonator arrays for weighted modulation.

[0076] Specifically, the optical path transmission diagram of the optical computing beam splitter output module is as follows: Figure 4 As shown.

[0077] The optical signal, after optical addition in the coupled waveguide, is transmitted to the last waveguide Bragg grating. This application utilizes the physical structural symmetry of the waveguide Bragg grating array to act as a waveguide demultiplexer on the optical signal output side. Optical signals matching the response band of the last waveguide Bragg grating are directly reflected to the corresponding output port. Mismatched optical signals pass through the last waveguide Bragg grating and are transmitted along the series waveguide to the penultimate waveguide Bragg grating for matched reflection or further transmission. This process continues until all n sets of optical signals are demultiplexed by their corresponding waveguide Bragg gratings and then guided to their respective n independent output ports.

[0078] Figure 5 This is a schematic diagram illustrating the matching principle between the passband of the waveguide Bragg grating and the resonant wavelength of the microring resonator, as provided in an embodiment of the present invention.

[0079] In terms of spectral distribution, the sharp peaks in the spectral response represent the resonance peaks of each microring resonator, where each resonator... At wavelength , , The resonant peak is exhibited at this location. The resonant wavelengths of these m microring resonators have a preset uniform spacing k. Fine wavelength division multiplexing based on microring resonators ensures that the system can perform convolution kernel weighted modulation on m independent wavelength channels within a single passband.

[0080] at the same time, Figure 5The spectral response in the Chinese square represents the reflection passband of the waveguide Bragg grating.

[0081] n waveguide Bragg gratings divide the broad spectrum into n independent grating passband windows. The center wavelength spacing between adjacent waveguide Bragg gratings is consistent with the free spectral range of the microring resonator, and the center wavelength of each waveguide Bragg grating increases by one free spectral range period in turn.

[0082] The optical reflection passband width of each waveguide Bragg grating is set to... Where k is a preset uniform spacing of the resonant wavelengths, through the above parameter configuration, the reflection passband of the i-th waveguide Bragg grating can completely cover the multiple resonant peaks of the i-th group of optical signals. Under this architecture, coarse wavelength division multiplexing based on waveguide Bragg gratings ensures that the system can effectively separate the resonant peaks of the same microring resonator at different periods in space. The calculation results of each group are reflected to the corresponding output port, thereby eliminating inter-channel crosstalk across the free spectral range.

[0083] This not only significantly reduces the number of microring resonators required, using fewer components and reducing chip area to achieve higher system integration, but also, since weighted modulation of the microring resonator array consumes power, sharing the same array of microring resonators with n groups of optical signals increases the computational load and speed by n times without increasing energy consumption, greatly improving computational speed and power while reducing losses and increasing integration.

[0084] The working process of this invention is described below: The input matrix X to be convolutionally processed is loaded into a total of There are n light wavelengths. These wavelengths are divided into n groups, and the i-th group contains wavelengths. to represents the data in the i-th row of the input matrix. These n sets of optical signals are injected into their respective input ports. to When the i-th group of optical signals encounters the i-th waveguide Bragg grating with a matching response band, the m wavelengths of that group are reflected and propagate into the through waveguide. The array contains m microring resonators. to The elements of the convolution kernel weight vector W were set using thermal tuning. to Because the j-th wavelength in each group ( to These all correspond to microring resonators in different free spectral ranges. The resonant wavelength allows n groups of wavelengths to share the same micro-ring array.

[0085] Subsequently, wavelength to Through the corresponding microring to The input data and weights are multiplied. The optical signals that have undergone multiplication are then summed in the coupled waveguide to generate the signal corresponding to the i-th element in the output vector Y.

[0086] Then, these signals are transmitted to the output end. Utilizing the structural symmetry, the i-th waveguide Bragg grating on the output side acts as a demultiplexer, reflecting and guiding the i-th group of signals to the output port. This system completes one parallel convolution computation cycle. It effectively achieves n-way parallel m-point convolution operations without increasing hardware power consumption, and the number of channels and convolution kernels can be flexibly adjusted.

[0087] This invention significantly improves the computing speed and the number of parallel channels compared to the traditional method of optical computing based on microring resonators. With proper design, the number of computing channels can be increased from 1 to n, while increasing computing power and speed by n times without increasing power consumption.

[0088] Furthermore, the computing power of traditional optical computing based on microring resonators is limited by the total number of wavelengths, which is confined to a single free spectral range (e.g., only m wavelengths can be used, allowing for only m calculations at a time). This new design cleverly utilizes multiple wavelengths from different free spectral ranges within the microring resonator array of the optical multiply-accumulate module simultaneously through an optical input combining module and an optical output splitting module. With the same number of microring resonators, this allows for m×n calculations at a time, significantly increasing computing power and the number of parallel computing channels.

[0089] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0090] For ease of description, directional terms such as "front, back, up, down, left, right," "horizontal, vertical, horizontal," and "top, bottom" generally indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings. These terms are used solely for the convenience of describing this application and simplifying the description. Unless otherwise stated, these directional terms do not indicate or imply that the device or element referred to must have a specific orientation or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the scope of protection of this application. The directional terms "inner" and "outer" refer to the inner or outer contours relative to the components themselves. For example, if a device in the drawings is inverted, a device described as "above" or "on top of" other devices or structures will subsequently be positioned as "below" or "under" other devices or structures. Thus, the exemplary term "above" can include both "above" and "below." The device may also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatial relative descriptions used herein are interpreted accordingly.

[0091] Unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, an electrical connection, or a communication connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0092] Unless otherwise expressly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0093] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0094] It should also be noted that the terms "one embodiment," "another embodiment," or "embodiment" used in this specification refer to specific features, structures, or characteristics described in connection with that embodiment, which are included in at least one embodiment described in the general description of this application. The appearance of the same expression in multiple places in the specification does not necessarily refer to the same embodiment. Furthermore, when a specific feature, structure, or characteristic is described in connection with any embodiment, the intention is to suggest that implementing such a feature, structure, or characteristic in conjunction with other embodiments also falls within the scope of this application.

[0095] 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.

[0096] It should also be noted that the above are merely preferred embodiments of this application and do not limit the scope of patent protection of this application. Any equivalent structural or procedural changes made using the content of this application’s specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of this application.

Claims

1. A multi-channel parallel optical computing chip combining coarse and fine wavelength division multiplexing, characterized in that, include An optical computing input beam combining module is used to receive multiple sets of input optical signals and wavelength-division multiplex each set of input optical signals to the corresponding computing optical path, wherein the multiple sets of input optical signals are used to indicate the input data to be convolutionally processed; The optical multiply-add module is used to receive the optical signal in the optical path of the operation, perform weighted modulation and optical addition according to the weight of the convolution kernel, and output an optical signal carrying the result of the convolution operation. The optical computing beam splitter output module is used to receive optical signals carrying the results of convolution operations, perform wavelet decomposition and multiplexing, and then guide them to the corresponding output port.

2. The multi-channel parallel optical computing chip combining coarse and fine wavelength division multiplexing according to claim 1, characterized in that: The optical computing input beam combining module and the optical computing beam splitting output module share the same waveguide Bragg grating array; The waveguide Bragg grating array comprises multiple waveguide Bragg gratings connected in series via optical waveguides. Utilizing the structural symmetry of the waveguide Bragg grating array, it serves as the optical computing input beam combining module on the optical signal input side and as the optical computing beam splitting output module on the optical signal output side.

3. The multi-channel parallel optical computing chip combining coarse and fine wavelength division multiplexing according to claim 2, characterized in that: The optical multiply-add module includes a through waveguide, a coupled waveguide, and a micro-ring resonator array; The microring resonator array includes multiple microring resonators, and the multiple microring resonators are optically coupled to the through waveguide and the coupled waveguide, respectively. Each of the microring resonators is equipped with a micro heater for thermally tuning the corresponding microring resonator to set the convolution kernel weight, thereby weighting the optical signal entering the microring resonator. The modulated optical signal is coupled into the coupling waveguide to complete the optical addition.

4. The multi-channel parallel optical computing chip combining coarse and fine wavelength division multiplexing according to claim 1 or 3, characterized in that: The first waveguide Bragg grating in the optical computing input beam combining module is connected to the through waveguide in the optical multiply-add module; the last waveguide Bragg grating in the optical computing beam splitting output module is connected to the coupled waveguide in the optical multiply-add module. The multiple input ports are respectively connected to the input side of each waveguide Bragg grating, and are used to inject the multiple sets of input optical signals; the multiple output ports are respectively connected to the reflection output side of each waveguide Bragg grating, and are used to output the optical signal after wave decomposition and multiplexing.

5. The multi-channel parallel optical computing chip combining coarse and fine wavelength division multiplexing according to claim 1, characterized in that: The aforementioned multiple sets of input optical signals comprise n sets of optical signals, and each set of optical signals comprises m optical wavelengths; The microring resonator array in the optical multiply-add module contains m microring resonators with preset uniform spacing and resonant wavelengths. The m optical wavelengths in each group of optical signals correspond to the resonant wavelengths of the microring resonator array in different free spectral ranges, so that the n groups of optical signals share the same group of microring resonator arrays for weighted modulation.

6. The multi-channel parallel optical computing chip combining coarse and fine wavelength division multiplexing according to claim 5, characterized in that: The center wavelength spacing between adjacent waveguide Bragg gratings in the waveguide Bragg grating array is consistent with the free spectral range of the microring resonator, and the center wavelength increases by one free spectral range in turn. The optical bandwidth of each waveguide Bragg grating is , where k is a preset uniform spacing of the resonant wavelength, such that the passband of the i-th waveguide Bragg grating covers the wavelength range of the i-th group of optical signals, so as to match the response of the entire free spectrum range covered by the corresponding microring resonator.

7. The multi-channel parallel optical computing chip combining coarse and fine wavelength division multiplexing according to claim 1, characterized in that: The chip performs highly parallel photonic convolution multiplication-addition operations, employing waveguide Bragg gratings to achieve coarse wavelength division multiplexing, enabling multiple sets of input optical signals to share the same microring resonator array. Therefore, the core physical process of the operation can be abstracted as the parallel dot product summation of multiple input vectors with fixed convolution kernel weights. The specific matrix operation formula is expressed as follows: Here, matrix X represents the input data to be convolved onto the optical signal, containing n sets of optical signals, each set containing m wavelength channels. Each wavelength constitutes Matrix X; the convolution kernel weight vector W represents the convolution kernel set by the microring resonator array; n groups of optical signals share the same array containing m microring resonators, therefore the transmittance set by the micro heater constitutes... The convolution kernel weight vector W; vector Y represents the result of the multiplication and addition operation between matrix X and the convolution kernel weight W; finally, the optical computing beam splitting output module outputs n independent optical signal results in parallel, i.e., one Column vectors.

8. The multi-channel parallel optical computing chip combining coarse and fine wavelength division multiplexing according to claim 1, characterized in that, The chip works as follows: Load the input matrix X into The i-th group of optical signals contains wavelengths. to , representing the i-th row of matrix X; n groups of optical signals are injected into the input port respectively. to ; When the i-th group of optical signals encounters the i-th waveguide Bragg grating with a matching response band, the m wavelengths of that group are reflected and propagate into the through waveguide; the array contains m microring resonators. to The elements of the convolution kernel weight vector W are set by thermal tuning. to ; because the j-th wavelength in each group to These all correspond to microring resonators in different free spectral ranges. The resonant wavelengths of n groups of wavelengths can therefore share the same set of microring arrays; wavelength to Through the corresponding microring to The input data and weights are multiplied. The optical signals from the multiplication operation are then summed in the coupled waveguide to generate the signal corresponding to the i-th element of the output vector Y. After being transmitted to the output, the i-th waveguide Bragg grating on the output side acts as a demultiplexer, utilizing structural symmetry to reflect and guide the i-th group of signals to the output port. This completes one parallel convolution computation cycle.

Images