62 results about "Hybrid computation" patented technology

The present disclosure provides a computer-implemented system for automated augmentation of differential equation models. The system stores differential equations representing mechanistic behavior of physical or computational processes and constructs a hybrid computational solver by embedding a trainable universal approximator with adjustable parameters into the differential equations, where approximator outputs augment time derivatives during numerical integration. An iterative training process adjusts parameters through numerical integration, monitors integration failures, assigns infinite penalty values to loss functions when failures occur, and computes gradients using automatic differentiation otherwise. The system computes sensitivity metrics via Jacobian matrix evaluation, classifies input / output subsets as significant based on threshold-exceeding sensitivity metrics, generates a reduced approximator operating on classified subsets, and replaces the universal approximator with the reduced version to create an optimized solver.

The invention provides a sparse and dense state calculation method, device and equipment and a readable storage medium, and the method comprises the steps: receiving an input data stream, analyzing the feature information of the input data stream, and marking a sensitivity level for the data stream; according to the sensitivity level of the data stream, a corresponding calculation strategy is called, and the calculation strategy at least comprises secret state calculation, hybrid calculation and plaintext calculation; and executing a calculation strategy for calculation associated with the data stream according to the calculation strategy corresponding to the data stream in AI training and / or AI reasoning. Through the technical scheme of the specification, the sensitivity level is marked for the data stream, and the secret state, mixing or plaintext calculation strategy is dynamically matched, so that the overall calculation overhead is remarkably reduced while the high-sensitivity data security is ensured, and the AI training and reasoning efficiency is improved.

The invention discloses a high-precision quantum sensing environment physical quantity real-time monitoring system and method, and belongs to the technical field of environment monitoring. The system comprises a quantum sensing front end, a signal excitation and regulation module, a quantum classical hybrid computing engine, an edge cloud collaborative transmission module and a terminal application module. The quantum sensing front end adopts a quantum entanglement sensing array integrated by NV color center diamond and a rubidium atom gas chamber to synchronously sense a magnetic field, temperature, pressure and vibration; the signal excitation and regulation module optimizes regulation parameters and inhibits noise through a quantum optimal control algorithm; a quantum-classical hybrid calculation engine is combined with a quantum neural network and classical multi-algorithm collaboration to realize efficient resolving and data fusion of quantum state features. According to the invention, the monitoring precision, the real-time response speed and the anti-interference capability are obviously improved, synchronous monitoring of multiple physical quantities is supported, and the system is suitable for various scenes such as industry, scientific research and deep space exploration, and has wide application value.

The invention provides an uncorrelated parallel machine scheduling method, system and device, a medium and a product. The method comprises the steps of obtaining a main problem and a sub-problem corresponding to a scheduling task; obtaining a binary representation model of the main question; the scheduling quantum computer analyzes the binary representation model of the main problem to obtain a plurality of optional scheduling strategies; analyzing the sub-problem based on the plurality of selectable scheduling strategies to obtain a plurality of candidate sub-problem feedback information; screening the plurality of candidate sub-problem feedback information to obtain target sub-problem feedback information, and performing a next iterative analysis process based on the target sub-problem feedback information to obtain a plurality of new selectable scheduling strategies; and obtaining a target scheduling strategy until an iteration ending condition is met. Through the quantum and classical hybrid computing system, the quality and generation efficiency of the scheduling strategy are improved.

The invention relates to the technical field of high-speed aircraft ground simulation and electromagnetic property calculation, and provides a high-speed aircraft electromagnetic calculation method and system based on a quantum classification algorithm. A variable component sub-classifier is used as an intelligent scheduling core to perform feature recognition and region division on a whole flow field, and a transmission matrix method and a WKB approximation method are adaptively called according to a division result to perform hybrid calculation, so that the technical problem that a traditional single calculation strategy is difficult to collaboratively optimize in precision and efficiency is solved; the method has the advantages that global computing resources can be distributed according to needs, and the overall solving speed is greatly increased while high simulation precision of the key area is guaranteed.

A device, method, and hybrid computing architecture for a Symbolic Photonic Logic Gate (SPLG) system that performs data processing primarily in the optical domain. The system comprises a photonic substrate, including silicon-photonics, indium phosphide, or similar optically active materials, patterned with optical waveguides and interferometric structures configured to implement logic operations through controlled optical interference, phase modulation, or non-linear optical effects. The SPLG system is configured to execute Boolean and symbolic logic functions without requiring intermediate optical-to-electrical signal conversion. In operation, information is encoded into one or more optical parameters including phase, polarization, wavelength, or amplitude, and processed within a Symbolic Optical Core optimized for linear algebraic operations such as matrix multiplication and vector transformation. The architecture further comprises an electronic control layer configured to initialize, configure, and monitor the photonic logic elements while allowing the majority of computational operations to occur within the optical domain, thereby reducing electrical interconnect congestion and resistive heating relative to fully electronic processors.

The invention provides a high-energy-efficiency digital-analog hybrid CMOS Boolean satisfiability problem solver chip which comprises a solver core, an SRAM peripheral read-write drive circuit and a parallel-to-serial read-out circuit, the SRAM peripheral read-write drive circuit is connected with a processor core and used for reading and writing data, the parallel-to-serial read-out circuit is used for reading a final solution, and the solver core comprises a 218 * 50 processing unit array. The processing units in each column are interconnected through charge sharing lines so as to complete accumulation operation in a charge domain. The processing units in each row are interconnected through TCAM matching lines, and the word and sentence evaluation operation is completed. In addition, a graph coloring circuit is also integrated for grouping Boolean variables. In each iteration, the Boolean variables in the same group can be updated at the same time, so that the parallel computing capability of the solver is improved. The energy efficiency is improved by using a digital-analog hybrid calculation mode, and the degree of parallelism is improved by using a packet updating strategy.

The invention belongs to the technical field related to information analysis, and discloses a flow graph processing method and processing system based on sub-graph hybrid computation.The processing method comprises the steps that an original flow graph in a graph database is divided into a plurality of sub-graphs with the vertex total weight distributed in a balanced mode; executing a graph updating operation on the sub-graphs; matching a corresponding graph calculation mode for each updated sub-graph, matching the corresponding updated sub-graph with a full-amount calculation mode when the proportion of the vertexes influenced by the updating operation in the updated sub-graphs exceeds a preset proportion, and otherwise, matching with an incremental calculation mode; extracting a dependency relationship between the sub-graphs; performing graph calculation on each updated sub-graph in parallel by adopting the matched graph calculation mode to respectively obtain an intermediate result of each sub-graph; and transmitting and updating the intermediate result of each sub-graph based on the dependency relationship, and outputting a global processing result of the flow graph. According to the processing mode, the locality characteristic of the graph can be fully utilized, the calculation overhead is reduced, and the accuracy of a calculation result is ensured.

A compute device includes one or more volatile memory (VM) dies and a non-volatile memory (NVM) die, stacked with the VM dies. The NVM die includes a wide-bit input / output (I / O) interface. Through-silicon vias (TSVs) are formed through the VM dies and the NVM die and interconnected through microbumps. A logic die is stacked with the VM dies and the NVM die. The wide-bit I / O interface is configured to handle data transfers over the TSVs to and from the VM dies, through the logic die, at one of a data rate of the VM dies or at a reduced rate compared to the data rate of the VM dies.

The invention discloses a storage and calculation integrated photon calculation system and method based on a phase change metasurface and holographic light field modulation, and aims to solve the problems that storage and calculation are separated, the dynamic modulation speed is low and the precision is limited in an existing photon calculation framework. The invention provides a hybrid computing architecture which takes a blue-ray disc or a phase change storage wafer as a high-density holographic weight storage medium and is combined with a high-speed spatial light modulator and a photoelectric nonlinear activation circuit. Through an area array parallel reading mechanism, a large-scale weight matrix is read at one time by using beam-expanded coherent light, light velocity matrix multiplication is completed, and interlayer fast switching of the deep neural network is realized in combination with a streaming rotation or light beam scanning technology. According to the invention, the non-volatility and multistage refractive index characteristics of the phase change material are utilized to realize high-density on-chip storage and zero-power-consumption lossless reading of large model weight, and the problems of a video memory wall and an energy consumption wall in large model reasoning are effectively solved.

The invention discloses a hybrid calculation method for generic genome assembly, which is characterized in that based on a hybrid calculation framework guided by de novo assembly and a reference genome, the reference genome depth is integrated into the whole process of generic genome assembly, and the generic genome assembly is obtained by utilizing telomeres of a sibling and / or same species to telomere reference genome and 10-20kbHiFi read length. A high quality replacement long sequence is generated to simulate and / or replace up to millions of bp ONT reads, achieving continuous high precision generic genome assembly with low computational power consumption. According to the invention, the developed hybrid computing framework and algorithm tool are combined with de novo assembly and a hybrid computing tool based on a reference assembly method, so that the existing reference genome from the same or related species can be efficiently utilized, and a high-quality substitutive long sequence can be generated by combining PacBio HiFi read length; the sequences and de novo assembly results are integrated to improve the assembly quality of a group scale data set, and reliable gene information resources are provided for research and application of generic genomes.

The invention discloses a dense matrix and sparse matrix efficient coupling parallel method. The method comprises the following steps: S1, carrying out feature analysis on an input dense matrix and sparse matrix; s2, carrying out adaptive partitioning on the dense matrix and the sparse matrix; s3, creating a uniform matrix descriptor for each matrix block; s4, constructing a calculation dependency graph; s5, executing parallel computing; according to the dense matrix and sparse matrix efficient coupling parallel method, through data rearrangement of self-adaptive hybrid storage and cache perception, the cache hit rate is greatly increased, the bottleneck of memory access mode conflicts in hybrid calculation is overcome, and meanwhile, based on an accurate prediction model and dynamic scheduling, the performance data in the running process are collected. The method has the advantages that combined optimization of computing load and communication is achieved, high load balance and resource utilization rate under large-scale parallel are guaranteed, redundant format conversion overhead is almost eliminated by the aid of unified descriptors and inert conversion strategies, a computing pipeline is smoother, and overall performance is improved remarkably.

There is provided a hybrid computing arrangement this is configured to implement a simulation of a chemical system, wherein the hybrid computing arrangement includes a combination of a classical computer coupled to a quantum computer, wherein the hybrid computing arrangement is configured in use to receive input data and to generate corresponding processed output data from the input data, wherein the computing arrangement is: (a) configured to receive information describing a chemical system in the input data: (b) configured to process the information describing the chemical system using a pre-entangler algorithm and a fixed circuit algorithm to generate a quantum ansatz defining initial values for a quantum circuit computation, and a Hamiltonian from which is generated a variational circuit algorithm: (c) configured to compute using the quantum ansatz and the variation circuit algorithm a corresponding quantum circuit to generate quantum computations results; and (d) configured to process the quantum computational results to generate the output data including information describing an electron orbital simulation of the chemical system.

The invention discloses a fusion calculation method of a quantum classical hybrid calculation system and a related device, and the method comprises the steps: carrying out the splitting processing of a task request applied by the quantum classical hybrid calculation system, so as to obtain calculation subtasks independently completed by quantum calculation, classical calculation and intelligent calculation; in response to the split calculation sub-task type, configuring the heterogeneous calculation resources to the corresponding calculation sub-tasks; the heterogeneous computing resources comprise quantum computing resources, classical computing resources and intelligent computing resources; the computing resources are evaluated and analyzed according to the resource evaluation and intelligent analysis module to obtain a corresponding scheduling decision basis; inserting the calculation subtasks into the corresponding task queues, and scheduling the task queues based on the scheduling decision basis, the user-defined scheduling strategy and the first linkage relationship of the calculation subtasks; according to the method, a classical and efficient fusion calculation technical scheme of quantum is provided, and an efficient solution is provided for a complex calculation problem through computing power complementation.

A firmware management operation. The firmware management operation includes providing an information handling system with a distributed BIOS; identifying a processor environment installed on an information handling system from a plurality of processor environments, the processor environment comprising processor corresponding to a processor architecture; and, performing a hybrid compute zone operation via the distributed BIOS, the hybrid compute zone operation causing a processor of the processor environment to enter into a low power hybrid compute mode of operation, the low power hybrid compute mode of operation controlling a graceful shutdown of the information handling system.

System (100), comprising: a working memory that stores components executable by a computer; and a processor that is operatively connected to the main memory, which executes computer-executable components, wherein the computer-executable components include: a patch identification component (116) that identifies a workload which belongs to a set of patches which belongs to a hybrid cloud-based data processing platform, wherein the workload has a first workload part for which a first subset of patches is to be performed, and a second workload part; a patch management component (114) that: a second subset of patches is determined, which is to be executed with respect to a part of the first workload part based on hazard assessments of patches of the first subset of patches within a maintenance window, wherein the first subset of patches includes the second subset of patches, a third subset of patches is determined to be executed with respect to the second workload part, wherein the set of patches comprises the first subset of patches, the second subset of patches, and the third subset of patches; and a patch execution component (118) that executes the second subset of patches with respect to the part of the first workload part within the maintenance window while the first workload part of the cloud-based hybrid computing platform is offline, and executes the third subset of patches with respect to the second workload part while the second workload part of the cloud-based hybrid computing platform is online.

The invention discloses a medicine packaging bag printing quality control system based on dynamic tagging. The system comprises a multi-physical field sensor array, a microfluidic surface analysis chip, a dual-mode tagging device, a self-adaptive optical modulator, a nanoscale positioning mechanical arm, a full-spectrum imaging system, a 3D surface scanning module, an edge-cloud hybrid computing platform, a high-precision time synchronization module, a fault prediction sensor, an embedded block chain node and an industrial 5G gateway. According to the method, through multi-dimensional base material characteristic perception, Transform modeling and PPO optimization, real-time tagging execution, full-spectrum quality detection, fault prediction and block chain tracing, high-definition tagging, accurate positioning and low rejection rate are realized. The system adapts to complex base materials and high-speed production lines, remarkably improves defect detection sensitivity and production continuity, reduces data tracing cost, conforms to supervision standards, supports green manufacturing, has cross-scene universality, is low in debugging and adaptation cost, and provides an efficient and intelligent solution for the medicine packaging industry.

The present disclosure belongs to the technical field of physical carrier synchronization, and relates to a trajectory prediction method and device, a storage medium and an electronic device. The method comprises the following steps: receiving first carrier movement data corresponding to a game carrier sent by a server, and determining second carrier movement data possessed by the game carrier at a current time; the first carrier movement data comprises a data sending time; if the current time is greater than the data sending time, the first carrier movement data and the second carrier movement data are calculated respectively, and a first carrier position corresponding to a first target time and a second carrier position corresponding to the current time are obtained; the first target time is different from the current time by a time interval; and the first carrier position and the second carrier position are mixedly calculated to predict a carrier position of the game carrier at the first target time. In the present disclosure, the accuracy of the predicted carrier position is increased.

The present disclosure provides an analog-digital hybrid computing method and a neuromorphic system using the same. The neuromorphic system includes: an input signal section that generates an input signal; a synapse section that includes synapse units that receive the input signal and flow a current according to a set weight, each synapse unit including a storage unit that is provided in a memory array including an input electrode line and an output electrode line and generates an output signal; a digital computing section that digitizes and sums the output signals; and a control section that specifies a digitization bit number for each output electrode line and stores a logic state in the storage unit; causes the input signal to be applied to each storage unit of each synapse unit through each input electrode line; generates, for each output electrode line, a sum of the currents flowing from the storage units of each synapse unit by the input signal as an output signal; and causes the digital computing section to digitize the output signals according to the bit number and calculate a sum of the digitized output signals.

A module optimizes active and inactive components and resources of a hybrid computing system to optimize a combination of private and public resources to minimize cost and maximize performance of the hybrid computing system. A learning model may analyze past usage and present usage metrics of one or more components with respect to performance criteria or cost criteria. Cost factors associated with components of the private system may be based on wear—the higher the wear the less desirable a component's use becomes due to lower reliability and higher warranty costs. When activating a component of the private system, a deactivation of a higher wear component may be delayed allowing time for a recently activated component to be integrated with the private system. A resource of a public system may be used while deactivation of a high wear component is delayed.

The application discloses a machine room operation and maintenance management automatic inspection system based on intelligence, and belongs to the technical field of intelligence, comprising a multi-source data fusion processing module, a hidden fault probability distribution module, a fault root cause analysis module, a dynamic self-adaptive baseline generation module, a multi-element strategy autonomous healing decision module, a collaborative control execution module and a feedback optimization module. The application constructs a full-link intelligent system of multi-source data fusion, quantum neural network prediction, graph neural network tracing, dynamic baseline generation, reinforcement learning decision, transactional scheduling execution and closed-loop optimization. The hidden fault recognition capability is improved through quantum-classical hybrid calculation. The root cause is accurately positioned in combination with a causal graph. An optimal repair scheme is generated based on a dynamic baseline and a reinforcement learning strategy library. Finally, the reliability of cross-system operation is ensured through transactional scheduling, and the fault prediction accuracy is significantly improved.

The application discloses an intelligent meal arrangement method, which adopts a hybrid computing architecture of "sequential iteration + parallel computing". Precise optimization of a first store provides high-quality "seed" dishes for a menu set; in the sequential iteration stage, subsequent stores are forced to preferentially select dishes from a historical preferred selection menu set; in the remaining store parallel optimization stage, for the remaining stores not participating in the sequential iteration, a parallel computing task pool is constructed based on the final preferred selection menu set of the sequential iteration, and multi-threading or a distributed computing framework is used to simultaneously solve the parallel computing task pool, so that meal arrangement results of the remaining stores are obtained. The intelligent meal arrangement method can quickly arrange meals for multiple stores and greatly improve the repetition rate of core dishes of the stores. The application further discloses a computer storage medium and an intelligent meal arrangement system.

The invention provides a reconfigurable mixed domain calculation annealing chip and application thereof, which comprises a processor core, an SRAM (Static Random Access Memory) peripheral read-write driving circuit and a spin state read-out circuit, wherein the SRAM peripheral read-write driving circuit is respectively connected with the processor core and is used for reading and writing data, and the spin state read-out circuit is used for reading spin state bit information. The processor core comprises three top layer modules, namely an analog domain calculation module, a digital domain calculation module and a master control circuit, and comprises a sign bit calculation unit circuit for analog calculation, a multiply-accumulate and calculation unit circuit, a current steering array circuit for realizing a random function, an adder tree circuit for digital calculation and a nonlinear probability approximation circuit; the coefficient storage array circuit is used for coupling coefficient storage, the spin storage circuit is used for spin state storage, and the interface and instruction peripheral control circuit is used for spin state connection. The energy efficiency is improved by using a digital-analog hybrid calculation mode, and the area efficiency is improved by using a storage and calculation integrated strategy.

The present application relates to the technical field of vehicle automatic driving, and more particularly to an automatic driving method and device based on expert hybrid model reinforcement learning and a medium. The method comprises the following steps: S1, constructing a multi-expert deep reinforcement learning architecture, extracting common features based on a shared backbone network and generating respective expert outputs by multiple independent experts; S2, implementing task-aware expert mixing based on a cross-attention mechanism soft gating network; S3, calculating the consistency degree of the expert outputs and adaptively adjusting the target expert activation entropy; S4, constructing a task-aware soft mixed expert deep reinforcement learning model based on a task-aware soft mixed expert deep reinforcement learning algorithm, training the model until the reinforcement learning model converges, obtaining the trained reinforcement learning model, receiving information about the environment around the vehicle, and outputting a corresponding driving decision by the reinforcement learning model.

The invention discloses an intelligent education analysis and execution system and method based on a multifunctional all-in-one machine, and belongs to the field of education technology and office automation. According to the system, on the basis of hardware of an existing printing, copying and scanning all-in-one machine, a hybrid computing architecture is innovatively adopted, and a local computing unit and a cloud collaboration unit are integrated. The core of the method is that dual-mode correction and dual-mode work are supported: AI intelligent correction and standard answer correction are supported in correction; in the working mode, seamless switching between offline and online is supported. The system can selectively and automatically correct the test papers in batches and accurately print marks on the original papers, meanwhile, a dynamic personal knowledge graph is constructed by analyzing answer data, and accordingly, a personalized learning path is generated and printed and practice is consolidated. According to the invention, general office equipment is upgraded into an intelligent education terminal which is applicable to all scenes and integrates automatic correction, process evaluation, learning condition analysis and personalized tutoring.

The invention discloses an FEM-SEA hybrid calculation method and device for full-frequency-domain NVH simulation analysis and a storage medium, and belongs to the field of computer aided design. The method comprises the following steps: calculating the modal density n (f) of a to-be-analyzed structure in a target frequency band; the frequency domain type (low frequency, intermediate frequency or high frequency) is automatically judged based on the value of a modal density criterion n (f) delta f; and according to a judgment result, a finite element method module, a statistical energy analysis method module or an FEM-SEA hybrid analysis module is automatically selected and called for simulation calculation. Wherein the FEM-SEA hybrid analysis is based on the direct mixing field reciprocity theorem, and accurate coupling between the deterministic subsystem and the statistical subsystem is realized by constructing an overall dynamic stiffness matrix Dtot, calculating an energy conversion coefficient and solving an energy balance equation. According to the method, full-frequency-domain automatic analysis of NVH simulation from low frequency to high frequency is achieved, dependence of method selection on artificial experience is avoided, and the precision and efficiency of medium-frequency band analysis are remarkably improved.

1. Name of the product in this design: Graphical User Interface for Cross-Scale Computing Workflow of Electronic Devices. 2. Purpose of this design: An electronic device. 3. The key design features of this product are its graphical user interface content. 4. The image or photo that best illustrates the design points: Interface change state diagram 2. 5. If there are no design points, omit the rear view; if there are no design points, omit the left view; if there are no design points, omit the right view; if there are no design points, omit the top view; if there are no design points, omit the bottom view. 6. Purpose of the graphical user interface: to provide a hybrid computing workflow platform that integrates quantum computing and classical simulation, supporting the automated construction and execution of cross-scale simulation computing tasks. 7. Other situations requiring explanation: The main view is the initial interface. In the main view, drag the required calculation item (a rectangle containing the calculation item name) from the left menu bar to the middle blank area to enter the interface change state diagram 1. In the interface change state diagram 1, connect the corresponding calculation nodes according to the required calculation process (the top is the calculation node name, and the bottom is a rectangle containing the relevant calculation content or result) to enter the interface change state diagram 2. Click the "Run" button in the interface change state diagram 2 to start the calculation.

This invention presents a unified framework for addressing complex, multi-domain challenges through adaptive design optimization, equation discovery, and hypothesis generation. Central to the framework is the Mathematical Sphere Framework (MSF), which employs advanced machine learning and hybrid computing to interrelate equation families across fields such as CFD, FEM, CAD, and PLM. MSF enables cross-domain optimization with transparent, physics-based representations, fostering continuous insight generation and system evolution. A neural-assisted system refines equations from experimental and real-world data, advancing theoretical models and design exploration. Hybrid computing combines classical preprocessing with quantum optimization to enhance computational efficiency. A digital twin provides real-time simulation and predictive analysis, while lifecycle adaptability dynamically optimizes parameters across product stages. By improving efficiency, scalability, and interoperability, this invention transforms engineering workflows, supports innovation, and fosters cross-industry applications, including aerospace, energy, and manufacturing, driving progress in an ever-evolving landscape.

The application discloses a kind of unified computing architecture and method based on quantum-classical ternary field, belong to quantum computing and computer architecture technical field.The architecture includes: quantum-classical unified field layer, for the computing resources of quantum state and classical state is mapped as a continuous ternary computing field;Ternary logic operation layer is connected with the quantum-classical unified field layer, using the three logic states of manifesting pole (+1), returning pole (-1) and operating central axis (0) to carry out unified logic operation to the data in the ternary computing field;Field balance control layer is connected with the ternary logic operation layer, and built-in single composite operation hardware module consisting of exponential operation, logarithmic operation and difference operation, for real-time monitoring the operation state deviation of the ternary computing field, and through closed loop feedback, the operation process converges to steady state.The application first unifies quantum state and classical state in a ternary field framework and carries out calculation, through the ternary collaborative mechanism of "manifesting pole execution, returning pole verification, operating central axis calibration", the seamless switching between quantum and classical two computing paradigms, collaborative operation and zeroing closed loop are realized, applicable to quantum-classical hybrid computing chip, all-optical computing system and next-generation general-purpose computing platform.

This invention discloses a graph neural network accelerator and device based on topology reconstruction and dynamic early retirement. The graph neural network accelerator includes an interconnected graph preprocessing module and an accelerator body. The graph preprocessing module includes a feature preprocessor, a pruner, a community divider, a topology optimizer, and an encoder to generate optimized graph structure data in ECSR format. The accelerator body includes an aggregation engine, a combination engine, a hierarchical storage unit, and a global configuration controller. The combination engine includes a confidence evaluation unit for evaluating node convergence confidence during graph neural network inference to support dynamic early retirement acceleration. This invention aims to address the low inference efficiency of general-purpose CPUs and GPUs due to the sparse and unstructured hybrid computing characteristics, as well as the shortcomings of existing graph neural network accelerators in topology adaptability, computational resource utilization, and model support, thereby improving the inference efficiency of graph neural network accelerators.