Hybrid integrated photonic computing chip system with multi-dimensional multiplexing

Abstract

The application discloses a multi-dimensional multiplexing hybrid integrated photonic computing core particle system, which comprises a multi-wavelength light source, a wavelength division demultiplexer, a hybrid integrated photonic chip, a filter array and a photodetector array, wherein the multi-wavelength light source encodes a problem to be solved and generates a multi-wavelength encoded optical signal containing the problem to be solved; the wavelength division demultiplexer fans out the multi-wavelength encoded optical signal to different channels and injects the multi-wavelength encoded optical signal into the hybrid integrated photonic chip; the hybrid integrated photonic chip loads weights on the encoded optical signals in the multiple channels to realize multiplication operation, and the processed optical signals enter a femtosecond laser direct writing chip to realize three-dimensional evolution of the optical signals and generate all-optical computing result signals; the filter array selectively filters the all-optical computing result signals containing two computing results, and the computing results are obtained through photoelectric conversion and amplitude determination of the photodetector array. The application realizes distributed computing to effectively solve a complex graph problem through multi-dimensional super-multiplexing of wavelength, three-dimensional space, amplitude and time and combining with optical interconnection technology, dividing a large-scale graph into manageable paragraphs through a three-dimensional femtosecond photonic chip architecture, and distributing the paragraphs to multiple photonic chips, and each chip processes a specific part of the graph.

Description

Technical Field

[0001] This invention relates to a technology in the field of photonic chips, specifically a multi-dimensional multiplexed hybrid integrated photonic computing chip system. Background Technology

[0002] Existing photonic integration platforms still face significant limitations in realizing advanced functions and complex systems: although silicon-based photonic chips have mature processes, they suffer from low electro-optic modulation efficiency and weak nonlinear optical effects; the high cost and complex processes of III-V materials restrict their large-scale application; and these platforms generally employ planar processes, limiting their capabilities in three-dimensional integration and system function expansion. Summary of the Invention

[0003] To address the aforementioned shortcomings of existing technologies, this invention proposes a multi-dimensional multiplexed hybrid integrated photonic computing chip system. By using multi-dimensional hyper-multiplexing of wavelength, three-dimensional space, amplitude, and time, combined with optical interconnect technology, a large-scale graph is divided into manageable segments through a three-dimensional femtosecond photonic chip architecture. These segments are then assigned to multiple photonic chips, with each chip processing a specific part of the graph, thereby achieving distributed computing to effectively solve complex graph problems.

[0004] This invention is achieved through the following technical solution:

[0005] This invention relates to a multi-dimensional multiplexing hybrid integrated photonic computing chip system, comprising: a multi-wavelength light source, a wave demultiplexer, a hybrid integrated photonic chip, a filter array, and a photodetector array, wherein: the multi-wavelength light source encodes the problem to be solved and generates a multi-wavelength encoded optical signal containing the problem to be processed; the wave demultiplexer fans out the multi-wavelength encoded optical signal to different channels and injects it into the hybrid integrated photonic chip; the hybrid integrated photonic chip applies weights to the encoded optical signals in multiple channels to achieve multiplication operations; the processed optical signal enters a femtosecond laser direct-write chip for three-dimensional evolution of the optical signal and generates an all-optical computation result signal; the filter array selectively filters the all-optical computation result signal containing two computation results, and the photodetector array performs photoelectric conversion and amplitude determination to obtain the computation result.

[0006] The multi-wavelength light source includes a broadband light source and a tunable filter, wherein the tunable filter maps the problem to be solved onto the amplitude of different wavelengths, selectively filters the continuous spectrum output by the broadband light source by adjusting the center wavelength and transmittance parameters, and encodes the problem information into discrete wavelength optical signals with different amplitudes.

[0007] The hybrid integrated photonic chip includes a high-speed optical signal modulation chip and a three-dimensional photonic chip integrated through end-face coupling. The high-speed optical signal modulation chip generates an optical signal carrying computational weights by electronically modulating and loading weight coefficients based on the input multi-wavelength optical signal. The three-dimensional photonic chip performs optical field evolution processing in a three-dimensional waveguide structure based on the modulated optical signal to generate the final all-optical computation result signal.

[0008] The aforementioned end-face coupling achieves mode field modulation between the two photonic chips through precise control of femtosecond laser direct writing parameters, thereby reducing coupling loss.

[0009] The high-speed optical signal modulation chip consists of several pairs of series modulators forming parallel branches, wherein each branch includes at least two modulators.

[0010] The modulator is a Mach-Zehnder interferometer, a semiconductor optical amplifier, a micro-ring modulator, an electro-absorption modulator, an acousto-optic modulator, or a liquid crystal modulator.

[0011] The high-speed optical signal modulation chip is made of at least one of the following materials: lithium niobate, silicon, silicon nitride, and III-V compounds.

[0012] The three-dimensional photonic chip is made of at least one of the following materials: fused silica, phosphate glass, borosilicate glass, rare earth-doped glass, etc.

[0013] When the problem to be processed is a #P-Complete problem, the corresponding three-dimensional photonic chip adopts the following structure:

[0014] The three-dimensional photonic chip includes: a first wavelength beam combiner, a three-dimensional beam splitter array, a second wavelength beam combiner, and a beam splitter connected in sequence. Specifically: the first wavelength beam combiner combines photons of two different wavelengths to simultaneously inject two different graphs into the chip, enabling parallel processing of more graphs by expanding the input of the combiner to N wavelengths; the three-dimensional beam splitter array maps the edge connections to a rearrangement of each vertex degree based on the characteristics of the input signal; the second wavelength beam combiner sums the number of light points in each layer to generate a related optical signal for the vertex degree; and the beam splitter performs specific calculations on the graph's structural information based on the processed optical signal to generate a final optical signal describing the graph's relationships.

[0015] The rearrangement refers to the following: the number of light spots in different layers corresponds to the degree of the vertices in the graph, generating light signals that reflect the relationships between the vertices.

[0016] The all-optical computation result signal is obtained in the following way:

[0017] Step 1: Encode the graph problem to be solved into multi-wavelength optical signals using a multi-wavelength light source;

[0018] The aforementioned multiple wavelengths refer to each wavelength representing an edge in the graph. For example, in a graph containing six nodes (A, B, C, D, E, F), there may be 15 edges (AB, AC, AD...EF), each edge corresponding to a specific wavelength (e.g., AB→λ1, AC→λ2...EF→λ). 15 When solving a given graph problem, if an edge exists in the graph, the multi-wavelength light source outputs a light signal of the corresponding wavelength. For example, if the graph to be solved contains edges AB, AD, CD, and EF, the multi-wavelength light source outputs wavelengths λ1, λ3, λ4, and λ5 respectively. 12 and λ 15 The optical signal.

[0019] Step 2: Use a dewavelength division multiplexer to separate the mixed optical signals output from the multi-wavelength light source into independent optical fiber channels according to wavelength. Each channel corresponds to one edge in the diagram.

[0020] Step 3: Inject the multi-wavelength optical signals from two different wavelength ranges into the corresponding ports of the hybrid integrated photonic computing chip according to the correspondence between wavelength and edges in the graph, thereby enabling the simultaneous solution of two different graph problems.

[0021] Step 4: The signal output from the hybrid integrated photonic computing chip is separated by a filter array with different center wavelengths, distinguishing the optical signals in two wavelength ranges and providing a clear signal distribution for subsequent processing.

[0022] Step 5: The filtered optical signal is converted into a photoelectric signal by a photodetector array and outputs an electrical signal with an amplitude corresponding to the degree value of each node in the graph to be solved.

[0023] Step 6: By loading the signal of the sub-graph space into the lithium niobate photonic chip portion of the hybrid integrated photonic computing chip, the subgraph of the graph to be solved is traversed quickly, and the degree sequence of all subgraphs is generated.

[0024] Step 7: By judging the generated subgraph degree sequence, the solution of the specific #P-Complete problem is completed, thereby efficiently outputting the final calculation result. Technical effect

[0025] This invention utilizes a hybrid integration technology combining a lithium niobate photonic chip and a glass 3D photonic chip to precisely control the parameters of the femtosecond laser direct writing process, thereby achieving mode field modulation between the two types of photonic chips, reducing coupling losses, and realizing high-density, high-efficiency optical fan-out of the lithium niobate photonic chip. Compared to existing technologies, this invention realizes a high-speed, large-scale hybrid integrated photonic general-purpose platform. Through customized design of the lithium niobate photonic chip and the femtosecond 3D photonic chip, it can realize high-performance on-chip photonic computing systems and highly complex on-chip optical quantum computing systems. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the system of the present invention;

[0027] Figure 2 This is a physical image of a hybrid photonics computing chip.

[0028] Figure 3 This is a schematic diagram illustrating the effect of an example. Detailed Implementation

[0029] like Figure 1 As shown, this embodiment relates to a multi-dimensional multiplexing hybrid integrated photonic computing chip system, including a multi-wavelength light source 100, a dewavelength division multiplexer 200, a hybrid integrated photonic chip 300, a filter array 400, and a photodetector array 500. Specifically: the multi-wavelength light source 100 encodes the problem to be solved, generating a multi-wavelength optical signal carrying the problem input information; the dewavelength division multiplexer 200 separates the single-channel multi-wavelength optical signal from the multi-wavelength light source according to wavelength, and fans it out to multiple independent channels before injecting it into the hybrid integrated photonic chip 300; the hybrid integrated photonic chip 300 loads weights onto the input optical signal and performs multi-dimensional parallel processing to achieve all-optical computation of the problem, generating a multi-wavelength computation result signal; the filter array 400 selectively filters the computation result signal in a specific band to extract the target wavelength signal; the photodetector array 500 converts the filtered optical signal into an electrical signal, and the system outputs the final computation result by determining the amplitude of the electrical signal.

[0030] like Figure 1 As shown, for a graph with 5 nodes and 10 edges, each edge corresponds to one of the 10 inputs of the system. A multi-wavelength light source 100 outputs 20 independent wavelengths, distributed across two wavelength windows. These wavelengths are then divided into independent wavelengths by a demultiplexer 200 and injected into a hybrid integrated photonic chip 300, corresponding to the two graphs to be solved, enabling parallel processing of multiple graphs. The hybrid integrated photonic system 300 processes the optical signals and then injects them into an optical fiber bandpass filter 400. A photodetector 500 detects the optical signals in the two wavelength windows and converts them into electrical signals for output. The amplitude of the output signal is proportional to the degree sequence of the input graph. The hybrid integrated photonic system 300 combines a high-speed lithium niobate photonic processor 310 with a three-dimensional photonic processor 320 to realize a high-speed, large-scale photonic computing system.

[0031] The hybrid integrated photonic chip 300 includes a high-speed optical signal modulation chip 310 and a three-dimensional photonic chip 320. The high-speed optical signal modulation chip 310 loads corresponding weight information according to the input optical signal, performs intensity or phase modulation of the optical signal, and generates an optical signal with calculated weights. The three-dimensional photonic chip 320 receives the modulated optical signal, realizes the three-dimensional evolution and calculation operation of the optical signal through a multi-layer optical structure, and finally generates a multi-wavelength optical signal reflecting the calculation results.

[0032] like Figure 2 As shown, the high-speed optical signal modulation chip 310 consists of several modulators connected in parallel, wherein each branch includes one modulator.

[0033] The modulator described uses a TFLN Mach-Zehnder modulator with a folded electrode structure to control high-speed optical signals. This design significantly reduces the half-wave voltage, enabling the modulator to directly interface with FPGA signals without additional amplification, thus avoiding the reduction in modulation speed and increase in power consumption that are usually caused by amplification.

[0034] like Figure 3 As shown in (a), this embodiment relates to an NP-Hard Graph problem handling method for the aforementioned hybrid integrated photonic computing chip system, specifically the mapping between the target graph and the 3D photonic chip. Taking a six-node graph as an example, such a graph can have a maximum of 15 edges, corresponding to 15 input ports on the chip. In this system, connections between nodes in the graph are represented by injecting photons into the corresponding input ports. Conversely, no connection between two nodes is represented by not injecting photons into the corresponding ports.

[0035] The three-dimensional photonic chip 320 for solving the six-node graph problem includes a first wavelength beam combiner 321, a three-dimensional beam splitter array 322, a second wavelength beam combiner 323, and a beam splitter 324 connected in sequence. The first wavelength beam combiner 321 combines photons of two different wavelengths, allowing two different graphs to be injected into the chip simultaneously. By expanding the input of the beam combiner from 2-in-1 to N-in-1, parallel processing of more graphs can be achieved, improving computational efficiency. The three-dimensional beam splitter array 322 maps the edge connections in the graph to a rearrangement of vertex degrees based on the input signal, where the number of light points in different layers corresponds to the vertex degree information, generating an optical signal representing the vertex connection relationship. The second wavelength beam combiner 323 sums the number of light points in each layer, generating a signal reflecting the vertex degree. After receiving the above signal, the beam splitter 324 further processes the graph's structural information, generating an optical signal describing the graph's characteristics or calculation results, providing support for subsequent calculations or output.

[0036] The first wavelength combiner 321 is a plurality of parallel two-in-one combiner structures.

[0037] The three-dimensional beam splitter 322 is a multi-dimensional arrangement of one-to-two beam splitting structures.

[0038] The second wavelength beam combiner 323 is a three-dimensional arrangement of five-in-one beam combining structures.

[0039] The beam splitter 324 is a plurality of parallel one-to-two beam splitting structures.

[0040] like Figure 3 As shown in (b)-(e), for a six-vertex graph, there are 15 connections between vertices (15 edges), which correspond to the 15 beam combiners at the chip input Part 1. When there is a connection between two vertices in the graph, photons are injected into the relevant port; otherwise, they are not. The evolution of different wavelength inputs in different Parts of the three-dimensional photonic chip is illustrated in the figure. Figure 3 As shown in (b), wavelengths from different input ports propagate in parallel within a waveguide after passing through a beam combiner, thus realizing the function of parallel computing; Figure 3 As shown in (c), two wavelengths of light propagating in the waveguide are uniformly split into two paths after passing through a beam splitter structure, with an intensity ratio of 1:1 between the two paths; Figure 3 As shown in (d), the beam combiner enables the addition function to allow multiple wavelengths to propagate in parallel; as... Figure 3 As shown in (e), the results of each sub-graph are obtained after multiple wavelengths propagating in parallel in the waveguide, after passing through a beam splitter structure and filters with different center wavelengths.

[0041] like Figure 3 As shown in (f), the output experimental results corresponding to the graph to be solved are obtained by performing analog-to-digital conversion on the light intensity and outputting the degree sequence of all subgraphs corresponding to the input graph to be solved.

[0042] like Figure 3 As shown in (g), based on Figure 3 As shown in (f), the system can effectively output all the minimum vertex cover solutions in the graph.

[0043] like Figure 3 As shown in (h), based on Figure 3 As shown in (f), the system can effectively output all the maximum clique solutions in the figure.

[0044] like Figure 3 As shown in (i), based on Figure 3 The experimental results shown in (h) correspond to all the maximum clique solutions in the graph to be solved. Based on this, considering the vertex weights, the solution to the weighted maximum clique problem is output.

[0045] The filter array 400 includes several parallel fiber bandpass filters, each filter having two different windows.

[0046] In a practical experiment, an AMD Artix-7 FPGA XC7A200T board was used as the external controller to drive the lithium niobate photonic chip at a modulation speed of 1 GHz, enabling it to traverse the complete solution space of a 6-vertex graph in 0.064 μs. This performance significantly surpasses that of existing laptops equipped with AMD Ryzen 5800H processors, which require 170 μs to complete the same task.

[0047] Compared to existing photonic integration based on single materials, this invention hybridizes a high-speed optical signal modulation photonic chip compatible with CMOS processes with a three-dimensional photonic glass chip fabricated using femtosecond laser direct writing technology. This provides high-speed optical signal modulation and complex photonic spatial evolution, compensating for the functional limitations of existing planar photonic chips. It achieves multi-dimensional hyper-multiplexing of wavelength, three-dimensional space, amplitude, and time, and, combined with optical interconnect technology, expands the system's scale and computing power, demonstrating a speed advantage over classical computers in solving the #P-Complete graph problem. This invention creates a scalable, large-scale, high-dimensional multiplexed photonic computing system architecture, laying the foundation for building all-optical computing systems that surpass classical electronic computing systems, marking a significant advancement in the large-scale application of photonic information processing technology.

[0048] The above-described specific implementations can be partially adjusted by those skilled in the art in different ways without departing from the principles and purpose of the present invention. The scope of protection of the present invention is defined by the claims and is not limited to the above-described specific implementations. All implementation schemes within the scope of the claims are bound by the present invention.

Claims

1. A multi-dimensional multiplexed hybrid integrated photonic computing chip system, characterized in that, include: The system comprises a multi-wavelength light source, a wave demultiplexer, a hybrid integrated photonic chip, a filter array, and a photodetector array. Specifically: the multi-wavelength light source encodes the problem to be solved and generates a multi-wavelength encoded optical signal containing the problem; the wave demultiplexer fans out the multi-wavelength encoded optical signal to different channels and injects it into the hybrid integrated photonic chip; the hybrid integrated photonic chip applies weights to the encoded optical signals in multiple channels to perform multiplication operations; the processed optical signal enters a femtosecond laser direct-write chip for three-dimensional evolution of the optical signal and generates an all-optical computation result signal; the filter array selectively filters the all-optical computation result signal containing two computation results; and the photodetector array performs photoelectric conversion and amplitude determination to obtain the computation result.

2. The multi-dimensional multiplexing hybrid integrated photonic computing chip system according to claim 1, characterized in that, The multi-wavelength light source includes a broadband light source and a tunable filter, wherein the tunable filter maps the problem to be solved onto the amplitude of different wavelengths, selectively filters the continuous spectrum output by the broadband light source by adjusting the center wavelength and transmittance parameters, and encodes the problem information into discrete wavelength optical signals with different amplitudes.

3. The multi-dimensional multiplexing hybrid integrated photonic computing chip system according to claim 1, characterized in that, The hybrid integrated photonic chip includes a high-speed optical signal modulation chip and a three-dimensional photonic chip integrated through end-face coupling. The high-speed optical signal modulation chip generates an optical signal carrying computational weights by electronically modulating and loading weight coefficients based on the input multi-wavelength optical signal. The three-dimensional photonic chip performs optical field evolution processing in a three-dimensional waveguide structure based on the modulated optical signal to generate the final all-optical computation result signal.

4. The multi-dimensional multiplexing hybrid integrated photonic computing chip system according to claim 3, characterized in that, The aforementioned end-face coupling achieves mode field modulation between the two photonic chips through precise control of femtosecond laser direct writing parameters, thereby reducing coupling loss.

5. The multi-dimensional multiplexing hybrid integrated photonic computing chip system according to claim 3, characterized in that, The high-speed optical signal modulation chip consists of several pairs of series modulators forming parallel branches, wherein each branch includes at least two modulators.

6. The multi-dimensional multiplexing hybrid integrated photonic computing chip system according to claim 3, characterized in that, When the problem to be processed is a #P-Complete problem, the corresponding three-dimensional photonic chip adopts the following structure: The three-dimensional photonic chip includes: a first wavelength beam combiner, a three-dimensional beam splitter array, a second wavelength beam combiner, and a beam splitter connected in sequence. Specifically: the first wavelength beam combiner combines photons of two different wavelengths to simultaneously inject two different graphs into the chip, enabling parallel processing of more graphs by expanding the input of the combiner to N wavelengths; the three-dimensional beam splitter array maps the edge connections to a rearrangement of each vertex degree based on the characteristics of the input signal; the second wavelength beam combiner sums the number of light points in each layer to generate a related optical signal for the vertex degree; and the beam splitter performs specific calculations on the graph's structural information based on the processed optical signal to generate a final optical signal describing the graph's relationships.

7. The multi-dimensional multiplexing hybrid integrated photonic computing chip system according to claim 6, characterized in that, The rearrangement refers to the following: the number of light spots in different layers corresponds to the degree of the vertices in the graph, generating light signals that reflect the relationships between the vertices.

8. The multi-dimensional multiplexing hybrid integrated photonic computing chip system according to any one of claims 1-7, characterized in that, The all-optical computation result signal is obtained in the following way: Step 1: Encode the graph problem to be solved into multi-wavelength optical signals using a multi-wavelength light source; Step 2: Use a dewavelength division multiplexer to separate the mixed optical signals output from the multi-wavelength light source into independent optical fiber channels according to wavelength. Each channel corresponds to one edge in the diagram. Step 3: Inject the multi-wavelength optical signals from two different wavelength ranges into the corresponding ports of the hybrid integrated photonic computing chip according to the correspondence between wavelength and edges in the graph, thereby enabling the simultaneous solution of two different graph problems; Step 4: The signal output from the hybrid integrated photonic computing chip is separated by a filter array with different center wavelengths to distinguish the optical signals in two wavelength ranges, providing a clear signal distribution for subsequent processing; Step 5: The filtered optical signal undergoes photoelectric conversion through a photodetector array and outputs an electrical signal with an amplitude corresponding to the degree value of each node in the graph to be solved. Step 6: By loading the signal of the sub-graph space into the lithium niobate photonic chip portion of the hybrid integrated photonic computing chip, the subgraph of the graph to be solved is traversed quickly, and the degree sequence of all subgraphs is generated; Step 7: By judging the generated subgraph degree sequence, the solution of the specific #P-Complete problem is completed, thereby efficiently outputting the final calculation result.

Images