Efficient data flow transceiving controller suitable for multi-chip storage and calculation system

By introducing a high-efficiency data stream transceiver controller into the multi-core in-memory computing system and using the send and receive controllers for dynamic data stream management, the path dependency problem in traditional bypass transmission is solved, achieving faster and more accurate data stream transmission and improved computing efficiency.

CN121009056APending Publication Date: 2025-11-2558TH RES INST OF CETC

Patent Information

Application Number
CN202511118570.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-08-11
Publication Date
2025-11-25

AI Technical Summary

Technical Problem

In traditional bypass transmission, bypass buffers and bypass switching rely on predefined paths, lacking dynamic adjustment capabilities and making it difficult to adapt to complex traffic changes, resulting in local bottlenecks and resource waste.

Method used

Design an efficient data stream transceiver controller suitable for multi-core in-memory computing systems. By setting up transmit and receive controllers around each core, data packets are processed using a configurable asynchronous FIFO, a data packetization unit, and an error correction code generation unit. Dynamic data redirection and buffer management are achieved through a bypass transmission channel.

Benefits of technology

It improves the transmission speed and accuracy of data streams between different cores, optimizes data stream scheduling, reduces invalid data transmission, reduces resource consumption, and improves the overall system throughput and computing efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121009056A_ABST
    Figure CN121009056A_ABST
Patent Text Reader

Abstract

The invention discloses an efficient data flow transceiving controller applicable to a multi-core storage and calculation system, which comprises two transmitting controllers and two receiving controllers, and the transmitting controllers and the receiving controllers are arranged on the peripheries of cores and are used for performing data packaging and transmitting and data flow judgment work; wherein the sending controller comprises a configurable asynchronous FIFO (First In First Out), a data packaging unit and an error correction code generating unit; the receiving controller comprises an error correction code decoding unit, a data packet unpacking unit and a microchip detection unit; the flit detection unit is used for detecting information in a head flit to judge whether data meet the calculation requirement of the current core particle or not, and if the data do not belong to the task range of the current core particle, the data are redirected to the sending controller through a bypass transmission mechanism. According to the application, the transmission speed of the data stream among different core particles is improved, and the data stream can be dynamically planned and transmitted according to the preset rule by constructing the efficient data transceiving controller.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of core particle design, and particularly relates to a high-efficiency data stream transceiving controller suitable for a multi-core particle storage and computing system. BACKGROUND

[0002] The multi-core particle storage and computing system is a novel computing system combining a multi-chip architecture and storage and computing integration technology. It realizes efficient data transmission and computing storage collaborative processing by integrating multiple core particles (chiplets) in a single package and optimizing the on-chip network and interconnection architecture. This architecture not only improves the computing density, but also effectively reduces the data movement overhead and reduces the storage access bottleneck, and is particularly suitable for artificial intelligence, image processing and high-performance computing fields. By reasonably designing the topology structure, optimizing the AXI / PCIe interconnection, and applying the network structure splitting technology, the multi-core particle storage and computing system can break through the limitations of the traditional von Neumann architecture and realize higher computing resource utilization.

[0003] When mapping a neural network in the storage and computing architecture, the characteristics of in-memory computing are utilized to directly store the weight matrix in the IMC crossbar array, so that the computing task can be efficiently completed in the storage unit, thereby avoiding high-power data movement and improving the computing throughput. In terms of data stream scheduling between computing units, the architecture optimizes the on-chip network (NoC) and the package network (NoP). Among them, the NoC is responsible for the data transmission between the computing units in the core particle, and the NoP realizes the cross-chip particle communication. In the current common storage and computing architecture, cross-chip particle data transmission depends on independent NoP transmitter (TX) and NoP receiver (RX) circuits, as well as a dedicated NoP router (Router) to perform data packet scheduling. The router adopts a preset routing mechanism to determine the target core particle of the data packet, and transmits the data to the TX / RX module through a dedicated port, thereby completing the receiving and sending operations of the data packet. In the current part of the efficient interconnection architecture, such as the Floret architecture based on the space-filling curve, the cross-chip particle data transmission mode adopts a one-way data stream, reduces unnecessary router ports and interconnection links through a double-port design, thereby reducing the interconnection complexity of the package layer, reducing resource occupation and optimizing power consumption.

[0004] Traditional hardware-level bypass transmission is a design strategy to optimize data flow, which bypasses the core processing unit through a dedicated mechanism to achieve efficient data transmission. It consists of three parts: bypass buffer, bypass switch and bypass routing. Among them, the bypass buffer sets up an independent buffer on the data path, so that data can be transmitted directly without passing through the main processing unit; the bypass switch plays a role in the on-chip network, allowing data to be transmitted directly from the input port to the target port without passing through the intermediate node; the bypass routing can bypass the congested node and adjust the data flow path.

[0005] In the Floret compute-in-memory architecture system described above, when mapping a neural network, a problem that a kernel cannot store the next convolution node may occur. Considering the maximum parameter load that a kernel can store alone, each kernel has 16 Tiles, each Tile has 40 PEs, and each PE has 128x128 ReRAM cells. Considering that all parameters in a convolution node with an input channel of 512 and a convolution kernel size of 3x3 are mapped to it, after quantization operation, each parameter is 8 bits, and after calculation, the node needs 1.8 kernels to store. At this time, although more bandwidth is needed for data stream transmission between kernels, it is still controllable. However, when mapping a cross-layer network (such as a ResNet residual network), the transmission of data flow in a residual block inside is relatively complex. If one-way transmission is still used in the architecture, some data flow may appear excessive and useless transmission in non-required kernels.

[0006] In traditional bypass transmission, the bypass buffer and bypass switch usually rely on predefined paths, lack dynamic adjustment capability, and are difficult to adapt to complex traffic changes. In high load conditions, local bottlenecks may still occur, affecting overall throughput. In addition, although the bypass mechanism reduces data processing overhead, it still requires additional hardware resources such as dedicated buffers and switch logic. SUMMARY

[0007] The present application aims to solve the problem that the bypass buffer and bypass switch in traditional bypass transmission usually rely on predefined paths and lack dynamic adjustment capability, and proposes a high-efficiency data stream transceiver controller suitable for a multi-kernel compute-in-memory system.

[0008] The present application provides a high-efficiency data stream transceiver controller suitable for a multi-kernel compute-in-memory system, comprising a plurality of kernels arranged in a chessboard pattern; each side of the four sides of each kernel is respectively provided with a sending controller or a receiving controller; one kernel comprises two sending controllers and two receiving controllers, and the sending controllers and receiving controllers are arranged on the periphery of the corresponding kernel for performing data packaging and sending and data flow judgment.

[0009] A bypass transmission channel is arranged between adjacent sending controllers or receiving controllers of the same chiplet;

[0010] A bypass transmission channel is arranged between adjacent sending controllers or receiving controllers of different chiplets;

[0011] The sending controller comprises a configurable asynchronous FIFO, a data packaging unit and an error correction code generation unit, the configurable asynchronous FIFO is used for storing and buffering the transmission data stream, the data packaging unit is used for packaging the micro-pieces of the stored data into data packets, and the error correction code generation unit is used for performing check code calculation on the data stream;

[0012] The receiving controller comprises an error correction code decoding unit, a data packet unpacking unit and a micro-piece detection unit, the error correction code decoding unit is used for detecting and correcting errors in the data stream transmission, the data packet unpacking unit is used for splitting the data packets into a micro-piece data stream, and the micro-piece detection unit is used for detecting information in the head micro-piece to determine whether the data meets the calculation requirements of the current chiplet, if the data does not belong to the task range of the current chiplet, the data is redirected to the sending controller through the bypass transmission mechanism.

[0013] Further, the micro-piece data stream contains a chiplet ID (CID), a current chiplet ID (NCID), a source node chiplet identifier (SCID), a destination node chiplet identifier (DCID), a source node router identifier (SRID) and a destination node router identifier (DRID).

[0014] Further, the sending controller and the receiving controller realize data interaction and forwarding through the bypass transmission channel;

[0015] The sending controller transmits data to the receiving controller adjacent to itself in the same chiplet through the bypass transmission channel, or transmits data to the receiving controller adjacent to itself and provided with a bypass transmission channel in the adjacent chiplet.

[0016] Further, when the input data stream data_in is transmitted to the first receiving controller, the first controller detects the relevant information of the received data packet according to a predefined analysis protocol, processes according to error correction code decoding, splits the verified data packet information after decoding as error-free transmission, and splits into multiple micro-piece data streams according to a preset condition; wherein the first receiving controller is the controller that is receiving data among the two receiving controllers;

[0017] After detecting the head micro-piece, it is judged whether the micro-piece information carried in the data is valid data required for calculation of the current chiplet; if the micro-piece information in the data packet matches the calculation requirements of the corresponding chiplet, the first receiving controller directly transmits the valid data to the corresponding chiplet for calculation and processing until the next head micro-piece appears.

[0018] If the microchip information contained in the data does not belong to the current core particle calculation range, the first receiving controller will redirect to the adjacent first sending controller for buffering through the bypass transmitter, waiting for sending;

[0019] Subsequently, the first sending controller continues to redistribute the data, packages multiple microchip data streams into a complete data packet, and then performs ECC encoding encryption on the data packet and transmits it to the target core particle.

[0020] Further, if the receiving controller determines that the data does not belong to the task range of the current core particle, the data is redirected to the buffer unit of the sending controller for temporary storage, rather than being stored in the dedicated bypass buffer.

[0021] Further, after the first receiving controller of the first core particle parses, it detects that the second data data_2 does not belong to the current core particle calculation task, and is forwarded to the adjacent first sending controller through the bypass transmitter. The first sending controller repackages the data stream according to the predefined logic and transmits it to the second receiving controller of the second core particle; wherein the first sending controller of the first core particle is adjacent to the second receiving controller of the second core particle.

[0022] At the same time, it is detected that the first data data_1 belongs to the core particle calculation task, and the first receiving controller of the first core particle directly transmits the first data to the integrated computing network on chip for calculation. After the calculation is completed, the result is transmitted to the second sending controller of the first core particle, and after the packaging processing, the output data stream data_out is transmitted to the next core particle again, completing the overall calculation and interaction of the data stream.

[0023] Further, each core particle independently performs data bypass transmission, and data forwarding is realized through multiple bypasses.

[0024] Further, the sending controller and the receiving controller correspond to different directions of the core particle respectively, and the two adjacent sides of the core particle are sending controllers and the other two sides are receiving controllers.

[0025] Compared with the prior art, the significant progress of the present application is that:

[0026] 1. The transmission speed of the data stream between different core particles is improved, and the data stream can be dynamically planned and transmitted according to the preset rules through the construction of an efficient data transceiver controller. This control mechanism introduces the core particle ID identifier to ensure that the data stream can be continuously compared and verified during transmission, not only optimizing the scheduling process of the data stream, but also ensuring that the data flows quickly and accurately between core particles, improving the overall throughput of the system. In addition, a reasonable scheduling strategy can reduce congestion during data transmission, so that the data stream remains efficient and stable in the multi-core architecture.

[0027] 2. Optimize the transmission of invalid data in the traditional architecture, introduce processing unit in the data receiving controller, judge the validity of data stream in real time, avoid useless data into the core grain to store and calculate unnecessarily. Compared with the traditional architecture, this improvement makes the data stream more accurately transmitted to the target module, avoids the performance loss caused by redundant data entering the processing unit.

[0028] 3. Resource optimization of bypass buffer, the reuse of buffer space in the data sending controller reduces the resources occupied by the bypass buffer. And no longer through the algorithm to select the path, but in each core grain separately bypass transmission, through multiple bypass to realize more simple data forwarding function, avoid the complex path calculation of traditional bypass transmission.

[0029] In order to more clearly illustrate the functional characteristics and structural parameters of the present application, the following will be further described in combination with the drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 is the NoP and NoC connection mode diagram in the common storage calculation architecture provided by the embodiment of the present application;

[0031] Figure 2 is the cross-chip interconnection mode schematic diagram in the Floret scheme provided by the embodiment of the present application;

[0032] Figure 3 is the data stream transceiver controller structure diagram provided by the embodiment of the present application;

[0033] Figure 4 is the bypass transmission scheme comparison diagram provided by the embodiment of the present application;

[0034] Figure 5 is the data stream transmission diagram provided by the embodiment of the present application;

[0035] Figure 6 is the implementation flowchart provided by the embodiment of the present application. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0037] The application replaces the receiving and sending controller of NoP in the original architecture. When the data stream is transmitted to the receiving controller of the chiplet, the data that is not processed by the chiplet is sent to the data sending controller through the bypass transmitter after detection, so as to ensure that the data stream that is not needed by the chiplet is not transmitted to the chiplet for useless transmission, and high-efficiency processing of the data stream in the chiplet is realized. Moreover, the traditional bypass transmission is modified to make it more suitable for the data receiving and sending controller.

[0038] The application provides a high-efficiency data stream receiving and sending controller suitable for a computing system. By adding a control architecture in the periphery of the chiplet, the data packet sending and data stream judgment work are performed, so as to ensure that the data stream that is not needed by the chiplet is not transmitted to the chiplet for useless transmission, but is transmitted to the chiplet where the data stream should be operated by the peripheral bypass transmitter.

[0039] The receiving and sending controller is also a partial improvement of the TX / RX of NoP, so the transmitted data stream flit data is newly defined to make the flit data stream transmission adapt to the receiving and sending controller. The transmitted data packet includes a head flit, a body flit and a tail flit, wherein the head flit is responsible for storing the ID information related to data transmission and routing, the body flit is responsible for storing the effective data load, and the tail flit is only used for marking the termination position of the data packet.

[0040] In the application, the data transmission across the chiplet is included, so the single source node ID and destination node ID in the head flit in the traditional scheme cannot completely satisfy the start and end positions of the data transmission, so the head flit includes a chiplet ID (Chiplet ID, CID) for indicating different chiplets, and a current chiplet ID (NCID) for ensuring the implementation transmission of the data and indicating the ID of the current chiplet where the data packet is located; a SCID (Sourece Chiplet ID) and a DCID (Destination Chiplet ID) are used to indicate the chiplet identification of the source node and the destination node of the data packet, and a SRID (Source Route ID) and a DRID (Destination Route ID) are used to indicate the router identification of the source node and the destination node of the data packet.

[0041] The sending controller is responsible for packaging and error correction code encoding of the data in the core grain, which is convenient for subsequent transmission of data stream between the core grains. The functional components of the sending controller include: the first part is a configurable asynchronous FIFO, which is used to store and buffer the transmission data stream, so that the data can be smoothly exchanged in the communication process of different rates; the second part is a data packaging unit, which is responsible for packaging the microfilm of stored data into a data packet according to the preset format for subsequent transmission; the third part is an error correction code generation unit, which calculates the check code of the data stream to ensure the integrity of the data in the cross-core transmission process and has error detection capability.

[0042] The receiving controller is responsible for error correction code decoding, data packet unpacking and data microfilm detection of the data transmitted in the core grain, which is convenient for subsequent correct transmission of data stream between the core grains. The functional components of the receiving controller include: the first part is an error correction code decoding unit, which is used to detect whether there is an error in the data stream during transmission and to correct the error; the second part is a data packet unpacking unit, which is responsible for splitting the transmitted data packet into multiple microfilm data streams, so that the microfilm data stream can be correctly used by the subsequent module; the third part is a microfilm detection unit, which is used to detect the information in the head microfilm, and judges whether the data meets the calculation requirements of the current core grain by comparing the destination core grain ID and the current core grain ID. If the corresponding microfilm data stream does not belong to the task range of the current core grain, the receiving controller will redirect the data to the sending controller through the bypass transmission mechanism and store it in the buffer unit, waiting for subsequent transmission and distribution.

[0043] In the embodiments of the present application, the north and west controllers are defined as data receiving controllers, and the south and east controllers are defined as data sending controllers, which is only one implementation. It should be noted that in actual definition, only the controllers at the two adjacent sides are positioned as data receiving controllers, and the other two sides are positioned as data sending controllers, and the specific direction needs to be limited. The controllers interact and forward data through the bypass transmission channel to ensure that the structure can better transmit data stream in the overall packaging level (such as the Mesh grid structure).

[0044] In a traditional bypass transmission scheme, after the data stream is subjected to decoding and unpacking, the micro-packet data stream is judged, and the bypass transmission operation is performed after the judgment. In the bypass transmission process, the data that needs to be bypassed is stored in the bypass buffer. After the buffering is completed, the bypass routing judgment and data bypass transmission operation are performed. In the bypass transmission design scheme of the present application, the data stream is no longer stored in the traditional dedicated bypass buffer, but is directly delivered to the next data sending controller after the micro-packet data stream judgment is completed, and is temporarily stored in the buffer unit of the sending controller. Therefore, the bypass transmission does not require an additional storage unit. This design reduces the resource occupation of the bypass buffer and optimizes the storage management of the data stream. Compared with the traditional bypass transmission mechanism in the entire network, the present design adopts the strategy of independent data bypass transmission of each core grain. This method allows the data to complete smaller granularity bypass processing within the core grain, realizes efficient data forwarding by using multiple bypasses, and simplifies the data flow path planning.

[0045] The data stream transmission case in the present application will be described in detail below.

[0046] Taking the input of the data_in data stream as an example, the method of the present application will be described in detail below.

[0047] When the input data stream data_in is transmitted to the west controller, the west controller detects the relevant information of the received data packet according to the predefined analysis protocol, processes according to the ECC (Error Correction Code) decoding, verifies the split data packet information after decoding as error-free data packet information, splits into multiple micro-packet data streams according to the preset conditions, analyzes the header information of the micro-packet, and detects the header micro-packet. After detecting the header micro-packet, it is judged whether the micro-packet information carried in the data is valid data required for calculation of the current core grain. If the micro-packet information in the data packet matches the calculation requirement of the current core grain, the west controller will directly transmit the valid data to the corresponding core grain for calculation and processing, until the next header micro-packet is detected.

[0048] If the micro-packet information contained in the data does not belong to the calculation range of the corresponding core grain, the west controller will redirect it to the south controller for buffering and waiting for sending through the bypass transmitter. Subsequently, the south controller continues to redistribute the data, packs multiple micro-packet data streams into a complete data packet, encodes and encrypts the data packet through ECC, and delivers it to the target core grain to ensure that the data can reach the core grain resource that truly needs to be calculated.

[0049] Take the input data stream data_in as an example, after the west controller of the core grain 1 parses, it is detected that data_2 in it does not belong to the current core grain computing task, and is accurately forwarded to the south controller through the bypass transmitter (the data stream corresponding to the black arrow tail wing in the figure), the south controller repackages the data stream according to the predefined logic, and transmits it to the north controller of the core grain 2, realizing cross-core grain data communication and cooperative computing. At the same time, it is detected that data_1 belongs to the core grain computing task, and the west controller directly transmits the data to the computing-on-network-on-chip for computing (the data stream corresponding to the black arrow white tail wing in the figure), and after the calculation is completed, the result is transmitted to the east controller of the core grain 1, and after the packaging processing, the output data stream data_out is transmitted to the next core grain again, completing the overall calculation and interaction of the data stream.

[0050] The key point of the application is to realize efficient transmission of cross-core grain data stream of the storage and computing system by constructing a data stream receiving and sending controller outside the core grain.

[0051] Efficient data stream receiving and sending control mechanism, by constructing a dynamically planned data receiving and sending controller, optimizing data stream scheduling, and improving the transmission speed between core grains.

[0052] Efficient data stream receiving and sending control mechanism, by constructing a dynamically planned data receiving and sending controller, optimizing data stream scheduling, and improving the transmission speed between core grains.

[0053] Optimization of bypass transmission, by reusing the buffer space to reduce the resource occupation of the bypass buffer area, and adopting a local bypass transmission mode instead of complex path calculation, so that the data forwarding process is more simple and efficient.

[0054] This optimization scheme improves the computing efficiency between core grains, reduces resource consumption, enhances the collaboration ability of cross-core grain data stream, and builds an efficient data stream receiving and sending controller through data stream analysis and splitting, intelligent bypass transmission and data stream redistribution.

[0055] Although embodiments of the application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made without departing from the principles and spirit of the application, and the scope of the application is defined by the appended claims and their equivalents.

Claims

1. A high-efficiency data flow transceiver controller suitable for multi-core grain computation system, characterized in that, The plurality of chiplets are arranged in a chessboard pattern; each of the four edges of each chiplet is respectively provided with a sending controller or a receiving controller; one chiplet includes two sending controllers and two receiving controllers, which are arranged on the periphery of the corresponding chiplet and used for performing data packet sending and data stream judgment; A bypass transmission channel is arranged between adjacent sending controllers or receiving controllers of the same chiplet; A bypass transmission channel is arranged between adjacent sending controllers or receiving controllers of different chiplets; The sending controller includes a configurable asynchronous FIFO, a data packet unit, and an error correction code generation unit; the configurable asynchronous FIFO is used for storing and buffering transmission data streams; the data packet unit is used for packaging the micro-pieces of the stored data into data packets; and the error correction code generation unit is used for performing error correction code calculation on the data stream. The receiving controller includes an error correction code decoding unit, a data packet unpacking unit, and a micro-piece detection unit; the error correction code decoding unit is used for detecting and correcting errors in the data stream transmission; the data packet unpacking unit is used for splitting the data packets into micro-piece data streams; and the micro-piece detection unit is used for detecting information in the head micro-piece to determine whether the data meets the calculation requirements of the current chiplet; if the data does not belong to the task range of the current chiplet, the data is redirected to the sending controller through the bypass transmission mechanism.

2. The high performance data flow transceiver controller of claim 1, wherein, The micro-piece data stream includes a chiplet ID (CID), a current chiplet ID (NCID), a source node chiplet ID (SCID), a destination node chiplet ID (DCID), a source node router ID (SRID), and a destination node router ID (DRID).

3. The high performance data flow transceiver controller of claim 1, wherein, The sending controller and the receiving controller realize data interaction and forwarding through the bypass transmission channel. The sending controller transmits data to the receiving controller adjacent to itself in the same chiplet through the bypass transmission channel, or to the receiving controller adjacent to itself and provided with a bypass transmission channel in the adjacent chiplet.

4. The high performance data flow transceiver controller of claim 3, wherein, When the input data stream data_in is transmitted to the first receiving controller, the first controller detects the relevant information of the received data packet according to a predefined analysis protocol, processes it according to error correction code decoding, splits it into multiple micro-piece data streams after verifying that it is a transmission error-free data packet information, and splits it into multiple micro-piece data streams according to a preset condition; wherein the first receiving controller is the controller that is receiving data among the two receiving controllers; After detecting the head micro-piece, it is judged whether the micro-piece information carried in the data is valid data required for calculation by the current chiplet; if the micro-piece information in the data packet matches the calculation requirements of the corresponding chiplet, the first receiving controller directly transmits the valid data to the corresponding chiplet for calculation and processing until the next head micro-piece appears; If the micro-piece information contained in the data does not belong to the calculation range of the current chiplet, the first receiving controller will be redirected to the adjacent first sending controller for buffering and waiting for sending through the bypass transmission device. Subsequently, the first sending controller continues to redistribute the data, packs the multiple micro-chip data streams into a complete data packet, and then performs ECC encryption on the data packet and transmits it to the target core.

5. The high performance data flow transceiver controller of claim 1, wherein, If the receiving controller determines that the data does not belong to the task of the core, the data is redirected to the buffer unit of the sending controller for temporary storage, rather than being stored through the dedicated bypass buffer.

6. The high performance data flow transceiver controller of claim 5, wherein, After the first receiving controller of the first core analyzes the data, it is detected that the second data data_2 does not belong to the current core computing task, and is forwarded to the adjacent first sending controller through the bypass transmitter, and the first sending controller repackages the data stream according to the predefined logic and transmits it to the second receiving controller of the second core; wherein the first sending controller of the first core and the second receiving controller of the second core are adjacent; At the same time, it is detected that the first data data_1 belongs to the core computing task, and the first receiving controller of the first core directly transmits the first data to the integrated computing on-chip network for calculation, and after the calculation is completed, the result is transmitted to the second sending controller of the first core, and after the packing processing, the output data stream data_out is transmitted to the next core again, completing the overall calculation and interaction of the data stream.

7. The high performance data flow transceiver controller of claim 1, wherein, Each core independently performs data bypass transmission, and data forwarding is achieved through multiple bypasses.

8. The high performance data flow transceiver controller of claim 1, wherein, The sending controller and the receiving controller correspond to different directions of the core respectively, and the two adjacent sides of the core are sending controllers and the other two sides are receiving controllers.

Citation Information

Patent Citations

  • Conversion interface of standard communication protocol and on-chip packet transport protocol for interconnection of multiple bare cores

    CN114756494A

  • Bus pipeline structure and chip for core-to-core interconnection

    CN115248796A

  • System and method for high speed packet transmission implementing dual transmit and receive pipelines

    US6901072B1

Cited By

  • First core grain, first chip, inter-core grain communication method and electronic equipment

    CN122173441A

  • First chip, first ferrite core, inter-chip communication method, and electronic equipment

    CN122173441B