CN116614433A - Artificial intelligent chip, data transmission method and data transmission system - Google Patents

Artificial intelligent chip, data transmission method and data transmission system Download PDF

Info

Publication number
CN116614433A
CN116614433A CN202310877938.0A CN202310877938A CN116614433A CN 116614433 A CN116614433 A CN 116614433A CN 202310877938 A CN202310877938 A CN 202310877938A CN 116614433 A CN116614433 A CN 116614433A
Authority
CN
China
Prior art keywords
data
chip
data packet
data transmission
artificial intelligent
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202310877938.0A
Other languages
Chinese (zh)
Other versions
CN116614433B (en
Inventor
李晓帅
王梦嘉
赵鹏程
王煇
严涵
华孙山
刘静雯
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Taichu Wuxi Electronic Technology Co ltd
Original Assignee
Taichu Wuxi Electronic Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Taichu Wuxi Electronic Technology Co ltd filed Critical Taichu Wuxi Electronic Technology Co ltd
Priority to CN202310877938.0A priority Critical patent/CN116614433B/en
Publication of CN116614433A publication Critical patent/CN116614433A/en
Application granted granted Critical
Publication of CN116614433B publication Critical patent/CN116614433B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L45/00—Routing or path finding of packets in data switching networks
    • H04L45/14—Routing performance; Theoretical aspects
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/16—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

The invention discloses an artificial intelligent chip, a data transmission method and a data transmission system, which relate to the technical field of chips, wherein the chip comprises: the control engine module is used for assembling first target data to be carried to obtain at least one first data packet and sending the at least one first data packet to the routing function module; wherein, the first data packet is added with a first target address; and the routing function module is used for determining at least one first communication interface in all the communication interfaces according to the routing strategy matched with the current data transmission mode and sending out a first data packet through the first communication interface. According to the technical scheme provided by the embodiment of the invention, the AI chip has a routing forwarding function, so that the AI chips interconnected in a diversified network mode can meet the data carrying requirements in different service scenes, the dependence on the data exchange chip when the data transmission system is constructed is avoided, and the data carrying speed of the constructed data transmission system is improved.

Description

Artificial intelligent chip, data transmission method and data transmission system
Technical Field
The present invention relates to the field of chip technologies, and in particular, to an artificial intelligent chip, a data transmission method, and a data transmission system.
Background
With the generation of multi-node massive parallel computing demands, high-speed interconnection communication between AI (Artificial Intelligence ) chips becomes an important research topic in the chip technical field.
The conventional interconnection scheme of AI chips is shown in fig. 1, and the AI chips under a central processing unit are connected through a Switch chip of a PCIe (peripheral component interconnect express, high-speed serial computer expansion bus standard) bus; as shown in FIG. 2, injettia (NVIDIA) also provides an interconnection scheme based on a dedicated switching chip (i.e., NV_switch chip); as shown in fig. 3, the chilies provide an 8-chip interconnection scheme based on an MLU-Link bridge card.
However, the interconnection scheme shown in fig. 1 is limited by the PCIe bus protocol and the PCIe topology, and the data bandwidth often cannot meet the actual load requirement; in the interconnection mode shown in fig. 2, a special conversion chip is required to be configured, so that higher hardware cost is required, and the interconnection scale of most chip systems cannot reach the above scale, and a chip function redundancy phenomenon often occurs; the interconnection scale shown in fig. 3 is only within 8 chips, and the expansibility is insufficient, so that the performance of a chip system is affected.
Disclosure of Invention
The invention provides an artificial intelligent chip, a data transmission method and a data transmission system, which are used for solving the problem of dependence on a data exchange chip when the artificial intelligent chip is used for constructing a large-scale network.
According to an aspect of the present invention, there is provided an artificial intelligence chip including: the system comprises a control engine module, a routing function module and a plurality of communication interfaces for interconnection with other hardware devices;
the control engine module is used for assembling first target data to be carried to obtain at least one first data packet and sending the at least one first data packet to the routing function module; wherein, the first data packet is added with a first target address;
the routing function module is used for determining at least one first communication interface in all communication interfaces according to a routing strategy matched with the current data transmission mode, and sending out the first data packet through the first communication interface.
According to another aspect of the present invention, there is provided a data transmission method applied to the artificial intelligence chip according to any embodiment of the present invention, including:
the control engine module assembles first target data to be carried to obtain at least one first data packet, and sends the at least one first data packet to the routing function module; wherein, the first data packet is added with a first target address;
The routing function module obtains at least one matched first communication interface through a congestion algorithm based on a cross switch matrix mode, and sends out the first data packet through the first communication interface.
According to another aspect of the present invention, there is provided a data transmission method applied to the artificial intelligence chip according to any embodiment of the present invention, including:
the control engine module assembles first target data to be carried to obtain at least one first data packet, and sends the at least one first data packet to the routing function module; wherein, the first data packet is added with a first target address;
the routing function module obtains a first transmission path according to the first target address and the address information of the current artificial intelligent chip based on a wireless network grid mode, and obtains a first communication interface corresponding to a first adjacent artificial intelligent chip in the first transmission path;
and the routing functional module sends the first data packet to the first adjacent artificial intelligent chip through the first communication interface.
According to another aspect of the present invention, there is provided a data transmission system including a plurality of artificial intelligence chips; the data transmission system is constructed based on a cross switch matrix mode; the artificial intelligent chips are interconnected in a cross switch matrix mode; the artificial intelligent chip is used for executing the data transmission method according to any embodiment of the invention.
According to another aspect of the present invention, there is provided a data transmission system including a plurality of artificial intelligence chips; the data transmission system is constructed based on a wireless network grid mode; the artificial intelligent chips are interconnected in a two-dimensional wireless network grid mode or in a three-dimensional wireless network grid mode; the artificial intelligent chip is used for executing the data transmission method according to any embodiment of the invention.
According to another aspect of the present invention, there is provided a data transmission device applied to the artificial intelligence chip according to any embodiment of the present invention, including:
the data packet acquisition module is configured in the control engine module and is used for assembling first target data to be carried to obtain at least one first data packet and sending the at least one first data packet to the routing function module; wherein, the first data packet is added with a first target address;
and the data packet sending module is configured in the routing function module and is used for acquiring at least one matched first communication interface through a congestion algorithm based on a cross switch matrix mode and sending the first data packet through the first communication interface.
According to another aspect of the present invention, there is provided a data transmission device applied to the artificial intelligence chip according to any embodiment of the present invention, including:
The data packet assembling module is configured on the control engine module and is used for assembling the first target data to be carried to obtain at least one first data packet and sending the at least one first data packet to the routing function module; wherein, the first data packet is added with a first target address;
the transmission path acquisition module is configured in the routing function module and is used for acquiring a first transmission path according to the first target address and the address information of the current artificial intelligent chip based on a wireless network grid mode and acquiring a first communication interface corresponding to a first adjacent artificial intelligent chip in the first transmission path;
and the sending operation execution module is configured in the routing function module and is used for sending the first data packet to the first adjacent artificial intelligent chip through the first communication interface.
According to another aspect of the present invention, there is provided a computer readable storage medium storing computer instructions for causing a processor to execute a data transmission method according to any one of the embodiments of the present invention.
According to the technical scheme, the control engine module assembles first target data to be carried to obtain at least one first data packet, and sends the at least one first data packet to the routing function module, the routing function module determines at least one first communication interface in all communication interfaces according to a routing strategy matched with a current data transmission mode, and sends the first data packet through the first communication interface, so that the AI chip has a routing forwarding function, and further the AI chips interconnected in a diversified network mode can meet data carrying requirements in different service scenes, dependence on data exchange chips when the data transmission system is built is avoided, the hardware structure of the data transmission system is simplified, and the data carrying speed of the built data transmission system is improved.
It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the invention or to delineate the scope of the invention. Other features of the present invention will become apparent from the description that follows.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic diagram of an AI chip interconnection method according to the background of the invention.
Fig. 2 is a schematic diagram of an AI chip interconnection method according to the background of the invention.
Fig. 3 is a schematic diagram of an AI chip interconnection method according to the background of the invention.
Fig. 4 is a schematic structural diagram of an AI chip according to a first embodiment of the present invention.
Fig. 5 is a schematic structural diagram of an AI chip according to a first embodiment of the present invention.
Fig. 6 is a schematic diagram of an AI chip direct connection mode according to a first embodiment of the present invention.
Fig. 7 is a schematic structural diagram of a data transmission system according to a second embodiment of the present invention.
Fig. 8 is a schematic structural diagram of yet another data transmission system according to the third embodiment of the present invention.
Fig. 9 is a schematic structural diagram of another data transmission system according to the third embodiment of the present invention.
Fig. 10 is a flowchart of a data transmission method according to a fourth embodiment of the present invention.
Fig. 11 is a flowchart of another method for transmitting data according to the fifth embodiment of the present invention.
Fig. 12 is a schematic structural diagram of a data transmission device according to a sixth embodiment of the present invention.
Fig. 13 is a schematic structural diagram of another data transmission device according to the seventh embodiment of the present invention.
Detailed Description
In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
Example 1
FIG. 4 is a block diagram of an artificial intelligence chip according to an embodiment of the present invention, the artificial intelligence chip including: a control engine module 100, a routing function module 200, and a plurality of communication interfaces 300 for interconnecting with other hardware devices; the control engine module 100 is configured to assemble first target data to be handled to obtain at least one first data packet, and send the at least one first data packet to the routing function module 200; wherein, the first data packet is attached with a first target address.
The control engine module 100 is a functional module in the AI chip for performing related operations such as data handling instruction analysis, data acquisition to be handled, and data packet generation; when the kernel of the AI chip acquires a data handling instruction sent by the central processing unit, forwarding the data handling instruction to the control engine module 100; the data handling instruction indicates that target data stored in the memory module of one AI chip is transferred to the memory module of another AI chip.
The data handling instructions include a source address, a destination address, and a data length; the source address represents a data initiator address and comprises a chip address where target data to be carried are located and a storage address of the target data in a storage module of the chip; the data length represents the data length of the target data to be handled; according to the source address and the data length, the complete data to be carried can be obtained; the target address represents the address of the data receiving party, including the chip address of the receiving party, and the storage address of the target data to be carried in the storage module of the chip of the receiving party.
As shown in fig. 5, when the control engine module 100 obtains a data handling instruction sent by the central processing unit, according to a source address and a data length, the control engine module obtains target data (i.e., first target data) to be handled from the storage module 500 through the network-on-chip module 400, assembles the first target data into one or more data packets (i.e., first data packets), and adds the target address (i.e., first target address) as additional information in each first data packet, and then sends all the first data packets to the routing function module 200; wherein the network-on-chip module 400 is further used for data transmission between the core 600 and the network-on-chip module 500.
According to the number of first data packets generated by the target data, the number of the first data packets is larger as the target data is larger, and the number of the first data packets is related to the size of the target data; the smaller the target data, the fewer the number of first data packets; in addition, the method is also related to the current data transmission mode of the AI chip, and the shorter the transmission path in the current data transmission mode is, the smaller the number of the first data packets is, so that the data carrying speed is increased; the longer the transmission path in the current data transmission mode, the larger the number of first data packets is to avoid transmission congestion caused by the oversized data packets.
The data transmission modes, that is, the interconnection mode of the current AI chip and other AI chips, may include a direct connection mode, a crossbar matrix mode and a wireless network mesh mode; a direct mode, i.e., the communication interface 300 of the current AI chip is directly connected with the communication interfaces 300 of other AI chips; the data transmission path is shorter in the direct connection mode, and the number of the first data packets may be set to a smaller value, for example, 1; the Crossbar mode, i.e., the Crossbar, is a connection of the communication interface 300 of the current AI chip with the communication interface of the data switching chip, i.e., the Switch chip, to communicate with other AI chips by means of the data switching chip; the data transmission path is longer in the crossbar matrix mode, and the number of the first data packets may be set to a larger value, for example, set to 4; the wireless network Mesh (Mesh) mode, i.e., the Mesh structure, is that the communication interface 300 of the current AI chip is connected with the communication interfaces 300 of other AI chips in a wireless network Mesh manner.
The data transmission mode may be configured when the current AI chip joins the system network, for example, a corresponding data transmission mode identification bit is configured in a storage module of each AI chip, and after the control engine module 100 reads the data transmission mode identification bit stored in the storage module 500 through the network-on-chip module 400, the control engine module sends the acquired data transmission mode to the routing function module 200, so as to complete the data transmission mode configuration of the AI chip; the data transmission mode can be obtained according to the data transmission mode identification information sent by the central processing unit, and the data transmission mode is sent to the routing function module 200, so that the data transmission mode configuration of the AI chip can be completed.
The routing function module 200 is configured to determine at least one first communication interface among all the communication interfaces 300 according to a routing policy matched with a current data transmission mode, and send out the first data packet through the first communication interface. In different data transmission modes, the routing function module 200 has different routing strategies; taking the example that the AI chips are interconnected in a direct connection mode, and each AI chip comprises 8 communication interfaces 300; FIG. 6 includes a schematic diagram of a structure of two AI chip interconnects, four AI chip interconnects, and eight AI chip interconnects, respectively; in an embodiment of the present invention, the communication interface 300 may include a serdes (serdes-Deserializer) interface, which is composed of a Serializer and a Deserializer, and which enables high-speed communication between AI chips through a serdes link.
In the direct connection mode, if it is determined that the receiver chip a corresponding to the first target address is directly connected to the current AI chip, the routing function module 200 selects one or more communication interfaces 300 among the communication interfaces connected to the receiver chip a by using the congestion algorithm as the first communication interface, and further sends out each first data packet through the first communication interface, and each first communication interface may send out one or more first data packets, so that the first target data to be carried is sent to the first target address.
If it is determined that the receiver chip a corresponding to the first destination address is not directly connected to the current AI chip, one or more of all the communication interfaces 300 currently connected to other AI chips may be selected as the first communication interfaces based on the congestion algorithm, and then each first data packet is sent out through the first communication interfaces, so that each first data packet is sent to the first destination address by means of forwarding of the other AI chips.
Optionally, in the embodiment of the present invention, the routing function module 200 is further configured to, in response to obtaining a second data packet sent by another artificial intelligent chip, determine, according to a second target address attached to the second data packet, whether the second target address is address information of the current artificial intelligent chip; and if the second target address is determined not to be the address information of the current artificial intelligent chip, forwarding the second data packet through the matched second communication interface according to the second target address.
Specifically, when the routing function module 200 obtains the second data packet sent by another AI chip through the communication interface 300, according to the additional target address (i.e., the second target address) in the second data packet, if it is determined that the second target address is not the address information of the current artificial intelligence chip, the forwarding operation of the second data packet is performed, so as to ensure smooth transmission of the second data packet; taking the data transmission mode as the direct connection mode as an example, if the routing function module 200 determines that the receiver chip B corresponding to the second target address is directly connected to the current AI chip, one of the communication interfaces connected to the receiver chip B is selected as the second communication interface through the congestion algorithm, and then the second data packet is sent out through the second communication interface, so that the second target data to be carried is sent to the second target address.
If it is determined that the receiver chip B corresponding to the second target address is not directly connected to the current AI chip, one of the communication interfaces 300 currently connected to other AI chips is selected as a second communication interface based on a congestion algorithm, and the second data packet is further sent out through the second communication interface, so that forwarding of the second data packet is continuously completed by means of the other AI chips.
Optionally, in the embodiment of the present invention, the routing function module 200 is further configured to send the second data packet to the control engine module 100 if it is determined that the second target address is address information of the current artificial intelligence chip; the control engine module 100 is further configured to store the second target data that is carried out in the second data packet to the second target address. Specifically, if the routing function module 200 determines that the second target address is the address information of the current artificial intelligent chip, a specific storage location in the storage module is determined according to the second target address, and then the remaining data after the additional second target address is removed in the second data packet, that is, the second target data after the handling is completed, is stored in the storage location, so that the handling of the second target data is completed.
According to the technical scheme, the control engine module assembles first target data to be carried to obtain at least one first data packet, and sends the at least one first data packet to the routing function module, the routing function module determines at least one first communication interface in all communication interfaces according to a routing strategy matched with a current data transmission mode, and sends the first data packet through the first communication interface, so that the AI chip has a routing forwarding function, and further the AI chips interconnected in a diversified network mode can meet data carrying requirements in different service scenes, dependence on data exchange chips when the data transmission system is built is avoided, the hardware structure of the data transmission system is simplified, and the data carrying speed of the built data transmission system is improved.
Example two
FIG. 7 is a block diagram of a data transmission system according to a second embodiment of the present invention, the system including a plurality of artificial intelligence chips; the data transmission system is constructed based on a cross switch matrix mode; the artificial intelligent chips are interconnected in a cross switch matrix mode; fig. 7 shows an example in which the data transmission system includes 12 AI chips, i.e., AI chip 0 through AI chip 7, and AI chips a through AI chip D.
Because each AI chip in the embodiment of the invention has a routing function, each AI chip can set the own identity information as a data exchange chip according to different functions executed by the AI chip in a data transmission system, namely the control engine module of the current AI chip does not execute the operation of reading data from a storage module or writing data into a reading module, and only can execute the data forwarding operation by the routing function module; the identity information of the control engine module can be set as a data read-write chip, namely, the control engine module of the current AI chip, and data reading operation and data writing operation can be performed.
In the data transmission system, the central processor can send an identity instruction to each AI chip to indicate the identity information of each AI chip in the current data transmission system; when the construction of the data transmission system is finished, configuring corresponding identification bits in a storage module of each AI chip so as to represent the identity information of the AI chip through the identification bits; after the control engine module in the AI chip acquires the identity instruction sent by the central processing unit or reads the identity mark stored in the storage module, the identity information of the current AI chip is determined. In fig. 7, AI chips 0 to 7 are data read/write chips, and AI chips a to D are data exchange chips.
If the control engine module determines that the identity information of the current artificial intelligent chip is a data read-write chip, after assembling target data to be carried to obtain a data packet and sending the data packet to the routing function module, the routing function module obtains at least one matched first communication interface through a congestion algorithm based on a cross switch matrix mode and sends the data packet through the first communication interface; for example, taking the current AI chip as AI chip 0 and the first target address as AI chip 7 as an example, the AI chip 0 has four communication interfaces to be connected to a data transmission system based on a cross-bar matrix mode, and a congestion algorithm may select a communication interface connected to AI chip a from the four communication interfaces as a first communication interface, and then sequentially send all first data packets to AI chip a through the first communication interface.
If the control engine module determines that the identity information of the current artificial intelligent chip is the data exchange chip, the routing function module judges whether the target address is the address information of the current artificial intelligent chip according to the target address attached to the data packet when acquiring the data packet sent by other artificial intelligent chips; if the routing function module determines that the target address is not the address information of the current artificial intelligent chip, forwarding the data packet through a communication interface matched with the target address; also taking the above technical solution as an example, if the current AI chip is AI chip a, as described above, AI chip a has already acquired the first target data sent by AI chip 0, and because AI chip a is connected to AI chip 7, AI chip a may forward the first target data to AI chip 7 through the communication interface connected to AI chip 7, so as to complete the handling of the first target data.
Particularly, if the identity information of the current AI chip is a data exchange chip, since the AI chip serving as the data exchange chip in the embodiment of the invention is an auxiliary operation chip with complete functions and still has the data calculation capability, when the control engine module obtains the transmission data through the routing function module, the control engine module can also execute corresponding data calculation so as to simplify the data processing flow of the data transmission system constructed based on the crossbar matrix and improve the data calculation efficiency.
Also taking fig. 7 as an example, in the conventional data transmission system, if the average value of data 0 in AI chip 0, data 1 in AI chip 1, data 2 in AI chip 2 and data 3 in AI chip 3 needs to be calculated, data 0 needs to be sent to AI chip 1 by AI chip 0, and after the average value of data 0 and data 1 is calculated by AI chip 1, the calculation result is sent to AI chip 2; the AI chip 2 calculates the average value of the received calculation result and the data 2, and then sends the calculation result to the AI chip 3; the AI chip 3 calculates the average value of the received calculation result and the data 3 to obtain a final calculation result, and then sends the final calculation result to the AI chip 1, the AI chip 2 and the AI chip 3; thus, the average value of the four data is obtained.
In the embodiment of the invention, only the AI chip 0, the AI chip 1, the AI chip 2 and the AI chip 3 are needed to respectively send the data 1, the data 2, the data 3 and the data 4 to the AI chip A, the AI chip A finishes the calculation of the average value of the four data, and then the calculation results are respectively sent to the AI chip 0, the AI chip 1, the AI chip 2 and the AI chip 3; obviously, in the data transmission system constructed based on the cross switch matrix in the embodiment of the invention, the identity information has the data calculation function of the AI chip of the data exchange chip, so that the data calculation efficiency of the system is greatly improved, and the data processing flow of the system is simplified.
Compared with the traditional technical scheme, a data exchange chip of a specific type (for example, a PCIe_switch chip in FIG. 1, an NV_switch chip in FIG. 2 and an MLU-Link bridge card in FIG. 3) is required to be arranged, and the complete data transmission system based on the cross Switch matrix can be formed by only using the AI chip of the same type in the embodiment of the invention, so that the chip composition type of the data transmission system is greatly simplified, and the hardware cost of the data transmission system is reduced; meanwhile, the identity information of each AI chip in the data transmission system can be modified by the mode that the central processing unit issues an identity instruction or changes the identity mark, so that the data transmission systems with different types and different functions can be constructed on the premise that the number and types of the AI chips and the connection modes among the AI chips are not changed, the functionality of the data transmission system is greatly enriched, and the convenience in constructing the data transmission system is improved.
Example III
Fig. 8 and fig. 9 are respectively two data transmission systems according to a third embodiment of the present invention, where each of the two data transmission systems includes a plurality of artificial intelligence chips; the data transmission system is constructed based on a wireless network grid mode; the individual artificial intelligence chips in fig. 8 are interconnected in a two-dimensional wireless network Mesh pattern (i.e., 2D Mesh); the artificial intelligence chips in fig. 9 are interconnected in a three-dimensional wireless network Mesh pattern (i.e., 3D Mesh); fig. 8 is an example of a data transmission system including 64 AI chips; fig. 9 is an example in which the data transmission system includes 256 AI chips.
In a data transmission system constructed based on a wireless network grid mode, the AI chip 0 is assumed to acquire a data carrying instruction sent by a central processing unit, and a target address to be carried is the AI chip 50; the control engine module of the AI chip 0 assembles the target data to be carried to obtain at least one data packet, and sends all the data packets to the routing function module; wherein, each data packet is added with a target address; in the wireless network grid mode, the addresses of each AI chip are arranged according to a certain rule, for example, the addresses are arranged in sequence of the first row and the last column or the first row and the last row; the routing function module may determine the relative location between AI chip 0 and AI chip 50 based on the target address and the address information of the current artificial intelligence chip in wireless network mesh mode.
The routing function module can obtain the optimal transmission path from the AI chip 0 to the AI chip 50 according to the relative positions, namely, the transmission path with the minimum number of hops; then, an AI chip adjacent to the AI chip 0 in the transmission path is acquired, and the adjacent AI chip is assumed to be the AI chip 3, and a data packet is sent to the AI chip 3 through a communication interface connected with the AI chip 3; in particular, adjacent AI chips, which are AI chips having a connection relationship with the current AI chip, may or may not be physically adjacent. In particular, the two-dimensional wireless network mesh mode differs from the three-dimensional wireless network mesh mode in that, when acquiring adjacent AI chips, the adjacent AI chips in the two-dimensional wireless network mesh mode may be one of the chips in four directions, i.e., one of the upper chip, the lower chip, the left chip, and the right chip; the AI chip in the three-dimensional wireless network mesh pattern may be one of the chips in six directions, i.e., one of the upper chip, the lower chip, the left chip, the right chip, the front chip, and the rear chip.
After the AI chip 3 obtains the data packet, it determines that the target address is not the address information of itself, then obtains the optimal transmission path according to the target address and the address information of itself, and further obtains the adjacent AI chip on the optimal transmission path, and sends the data packet to the adjacent AI chip, so as to realize the skip forwarding of the data packet through the forwarding of multiple AI chips.
Specifically, when the AI chip 0 sends out a data packet, the AI chip 3 can send a transmission path to the AI chip 3 synchronously, so that after the AI chip 3 acquires the data packet, the AI chip 3 can directly determine an adjacent AI chip based on the transmission path without calculating the transmission path again, and meanwhile, intermediate forwarding chips such as the AI chip 3 and the like send out the data forwarding path during each forwarding, so that the data carrying efficiency is improved; in addition, when the AI chip acquires the data forwarding path sent by the upstream AI chip, the AI chip can delete the address information of the AI chip in the data forwarding path, only the address information of the rest AI chips in the data forwarding path is reserved, so that the updating of the data forwarding path is completed, and the data volume of the transmission path sent each time is reduced; meanwhile, each AI chip in the transmission path is also convenient to directly execute data forwarding operation based on the first address information in the address information of the remaining AI chips, so that the data carrying efficiency is further improved.
Compared with the traditional technical scheme, in the data transmission system constructed based on the wireless network grid mode, the matched data exchange chips are required to be used for completing data interaction among all the AI chips, and obviously, under the condition that the total number of the chips is the same, the data transmission system in the embodiment of the invention does not need to introduce the data exchange chips, and each component chip of the system can be used for completing data forwarding, so that the chip component type of the data transmission system is simplified, the hardware cost of the constructed data transmission system is reduced, the functionality of the data transmission system is further enriched, and the convenience in constructing the data transmission system is improved.
Example IV
Fig. 10 is a flowchart of a data transmission method according to the fourth embodiment of the present invention, where the method may be performed by the data transmission device according to the sixth embodiment, and the data transmission device may be implemented in hardware and/or software, and the data transmission device is configured in the artificial intelligence chip according to the first embodiment. As shown in fig. 10, the method includes:
s401, the control engine module assembles first target data to be carried to obtain at least one first data packet, and sends the at least one first data packet to the routing function module; wherein, the first data packet is attached with a first target address.
S402, the routing function module obtains at least one matched first communication interface through a congestion algorithm based on a cross switch matrix mode, and sends out the first data packet through the first communication interface.
According to the technical scheme, the control engine module assembles first target data to be carried to obtain at least one first data packet, and sends the at least one first data packet to the routing function module, the routing function module obtains at least one matched first communication interface through a congestion algorithm based on a cross switch matrix mode, and sends the first data packet through the first communication interface, so that the AI chip has a routing forwarding function, and the AI chips interconnected in the cross switch matrix mode can meet data carrying requirements in different service scenes, so that dependence on data exchange chips in the process of constructing a data transmission system is avoided, the hardware structure of the data transmission system is simplified, and the data carrying speed of the constructed data transmission system is improved.
Example five
Fig. 11 is a flowchart of a data transmission method according to a fifth embodiment of the present invention, where the method may be performed by the data transmission device according to the seventh embodiment, and the data transmission device may be implemented in hardware and/or software, and the data transmission device is configured in the artificial intelligence chip according to the first embodiment. As shown in fig. 11, the method includes:
s501, the control engine module assembles first target data to be carried to obtain at least one first data packet, and sends the at least one first data packet to the routing function module; wherein, the first data packet is attached with a first target address.
S502, the routing function module obtains a first transmission path according to the first target address and the address information of the current artificial intelligent chip based on a wireless network grid mode, and obtains a first communication interface corresponding to a first adjacent artificial intelligent chip in the first transmission path.
S503, the routing function module sends the first data packet to the first adjacent artificial intelligent chip through the first communication interface.
According to the technical scheme, the control engine module assembles first target data to be carried to obtain at least one first data packet, and sends the at least one first data packet to the routing function module, the routing function module obtains a first transmission path based on a wireless network grid mode according to a first target address and address information of a current artificial intelligent chip, and obtains a first communication interface corresponding to a first adjacent artificial intelligent chip in the first transmission path, and further sends the first data packet to the first adjacent artificial intelligent chip through the first communication interface, so that the AI chip has a routing forwarding function, and the AI chips interconnected in the wireless network grid mode can meet data carrying requirements in different service scenes, avoid dependence on data exchange chips when the data transmission system is built, simplify a hardware structure of the data transmission system, and improve data carrying speed of the built data transmission system.
Example six
Fig. 12 is a block diagram of a data transmission device according to embodiment 6 of the present invention, where the device specifically includes:
the data packet obtaining module 601 is configured in the control engine module, and is configured to assemble first target data to be handled to obtain at least one first data packet, and send the at least one first data packet to the routing function module; wherein, the first data packet is added with a first target address;
the data packet sending module 602 is configured to obtain, based on a crossbar matrix mode, at least one matched first communication interface through a congestion algorithm, and send the first data packet through the first communication interface.
According to the technical scheme, the control engine module assembles first target data to be carried to obtain at least one first data packet, and sends the at least one first data packet to the routing function module, the routing function module obtains at least one matched first communication interface through a congestion algorithm based on a cross switch matrix mode, and sends the first data packet through the first communication interface, so that the AI chip has a routing forwarding function, and the AI chips interconnected in the cross switch matrix mode can meet data carrying requirements in different service scenes, so that dependence on data exchange chips in the process of constructing a data transmission system is avoided, the hardware structure of the data transmission system is simplified, and the data carrying speed of the constructed data transmission system is improved.
Optionally, the data transmission device further includes:
the address information judging module is configured in the routing functional module and is used for responding to the acquired second data packet sent by other artificial intelligent chips and judging whether the second target address is the address information of the current artificial intelligent chip according to a second target address attached to the second data packet;
and the data forwarding module is configured in the routing function module and is used for forwarding the second data packet through a second communication interface matched with the second target address if the second target address is determined not to be the address information of the current artificial intelligent chip and the current data transmission mode is determined to be a cross switch matrix mode.
The device can execute the data transmission method provided by any embodiment IV of the invention, and has the corresponding functional modules and beneficial effects of the execution method. Technical details not described in detail in the present embodiment can be referred to the data transmission method provided in the fourth embodiment of the present invention.
Example seven
Fig. 13 is a block diagram of a data transmission device according to a seventh embodiment of the present invention, where the device specifically includes:
the data packet assembling module 701 is configured in the control engine module, and is configured to assemble the first target data to be handled to obtain at least one first data packet, and send the at least one first data packet to the routing function module; wherein, the first data packet is added with a first target address;
The transmission path obtaining module 702 is configured to obtain a first transmission path according to the first target address and address information of the current artificial intelligent chip based on the wireless network mesh mode, and obtain a first communication interface corresponding to a first adjacent artificial intelligent chip in the first transmission path;
the sending operation executing module 703 is configured in the routing function module, and is configured to send the first data packet to the first neighboring artificial intelligence chip through the first communication interface.
According to the technical scheme, the control engine module assembles first target data to be carried to obtain at least one first data packet, and sends the at least one first data packet to the routing function module, the routing function module obtains a first transmission path based on a wireless network grid mode according to a first target address and address information of a current artificial intelligent chip, and obtains a first communication interface corresponding to a first adjacent artificial intelligent chip in the first transmission path, and further sends the first data packet to the first adjacent artificial intelligent chip through the first communication interface, so that the AI chip has a routing forwarding function, and the AI chips interconnected in the wireless network grid mode can meet data carrying requirements in different service scenes, avoid dependence on data exchange chips when the data transmission system is built, simplify a hardware structure of the data transmission system, and improve data carrying speed of the built data transmission system.
Optionally, the data transmission device further includes:
the judging and executing module is configured in the routing functional module and is used for responding to the acquired second data packet sent by other artificial intelligent chips and judging whether the second target address is the address information of the current artificial intelligent chip according to a second target address attached to the second data packet;
the path acquisition execution module is configured to the routing function module and is used for acquiring a second transmission path according to the second target address and the address information of the current artificial intelligent chip and acquiring a second communication interface corresponding to a second adjacent artificial intelligent chip in the second transmission path if the second target address is determined not to be the address information of the current artificial intelligent chip and the current data transmission mode is determined to be a wireless network grid mode;
and the forwarding execution module is configured in the routing function module and is used for forwarding the second data packet to the second adjacent artificial intelligent chip through the second communication interface.
Optionally, the data transmission device further includes:
the sending operation executing module 703 is specifically configured to send the first data packet and the first transmission path to the first neighboring artificial intelligence chip through the first communication interface, so that the first neighboring artificial intelligence chip executes a transmission operation of the first data packet based on the first transmission path.
The device can execute the data transmission method provided by the fifth embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method. Technical details not described in detail in this embodiment may be referred to the data transmission method provided in the fifth embodiment of the present invention.
Example eight
In some embodiments, the data transmission method may be implemented as a computer program tangibly embodied on a computer-readable storage medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and/or installed onto the heterogeneous hardware accelerator via the ROM and/or the communication unit. One or more of the steps of the data transmission method described above may be performed when the computer program is loaded into RAM and executed by a processor. Alternatively, in other embodiments, the processor may be configured to perform the data transmission method in any other suitable way (e.g., by means of firmware).
Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit systems, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems On Chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.
A computer program for carrying out methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions/acts specified in the flowchart and/or block diagram block or blocks to be implemented. The computer program may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.
In the context of the present invention, a computer-readable storage medium may be a tangible medium that can contain, or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer readable storage medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, the computer readable storage medium may be a machine readable signal medium. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
To provide for interaction with a user, the systems and techniques described here can be implemented on a heterogeneous hardware accelerator having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and pointing device (e.g., a mouse or a trackball) through which a user can provide input to the heterogeneous hardware accelerator. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.
The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), blockchain networks, and the internet.
The computing system may include clients and servers. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so that the defects of high management difficulty and weak service expansibility in the traditional physical hosts and VPS service are overcome.
It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps described in the present invention may be performed in parallel, sequentially, or in a different order, so long as the desired results of the technical solution of the present invention are achieved, and the present invention is not limited herein.
The above embodiments do not limit the scope of the present invention. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of the present invention.

Claims (10)

1. An artificial intelligence chip, comprising: the system comprises a control engine module, a routing function module and a plurality of communication interfaces for interconnection with other hardware devices;
the control engine module is used for assembling first target data to be carried to obtain at least one first data packet and sending the at least one first data packet to the routing function module; wherein, the first data packet is added with a first target address;
the routing function module is used for determining at least one first communication interface in all communication interfaces according to a routing strategy matched with the current data transmission mode, and sending out the first data packet through the first communication interface.
2. The artificial intelligent chip of claim 1, wherein the routing function module is further configured to, in response to obtaining a second data packet sent by another artificial intelligent chip, determine, according to a second destination address attached to the second data packet, whether the second destination address is address information of a current artificial intelligent chip; and if the second target address is determined not to be the address information of the current artificial intelligent chip, forwarding the second data packet through the matched second communication interface according to the second target address.
3. The artificial intelligence chip of claim 2, wherein the routing function module is further configured to send the second data packet to the control engine module if it is determined that the second destination address is address information of a current artificial intelligence chip;
and the control engine module is also used for storing the second target data which is carried out in the second data packet to the second target address.
4. A data transmission method applied to the artificial intelligent chip as claimed in any one of claims 1 to 3, comprising:
the control engine module assembles first target data to be carried to obtain at least one first data packet, and sends the at least one first data packet to the routing function module; wherein, the first data packet is added with a first target address;
the routing function module obtains at least one matched first communication interface through a congestion algorithm based on a cross switch matrix mode, and sends out the first data packet through the first communication interface.
5. The data transmission method according to claim 4, characterized in that the data transmission method further comprises:
the routing function module responds to obtaining a second data packet sent by other artificial intelligent chips, and judges whether the second target address is the address information of the current artificial intelligent chip according to a second target address attached to the second data packet;
And if the routing function module determines that the second target address is not the address information of the current artificial intelligent chip and determines that the current data transmission mode is a cross switch matrix mode, forwarding the second data packet through a second communication interface matched with the second target address.
6. A data transmission method applied to the artificial intelligent chip as claimed in any one of claims 1 to 3, comprising:
the control engine module assembles first target data to be carried to obtain at least one first data packet, and sends the at least one first data packet to the routing function module; wherein, the first data packet is added with a first target address;
the routing function module obtains a first transmission path according to the first target address and the address information of the current artificial intelligent chip based on a wireless network grid mode, and obtains a first communication interface corresponding to a first adjacent artificial intelligent chip in the first transmission path;
and the routing functional module sends the first data packet to the first adjacent artificial intelligent chip through the first communication interface.
7. The data transmission method according to claim 6, characterized in that the data transmission method further comprises:
The routing function module responds to obtaining a second data packet sent by other artificial intelligent chips, and judges whether the second target address is the address information of the current artificial intelligent chip according to a second target address attached to the second data packet;
if the routing function module determines that the second target address is not the address information of the current artificial intelligent chip, and determines that the current data transmission mode is a wireless network grid mode, acquiring a second transmission path according to the second target address and the address information of the current artificial intelligent chip, and acquiring a second communication interface corresponding to a second adjacent artificial intelligent chip in the second transmission path;
and the routing functional module forwards the second data packet to the second adjacent artificial intelligent chip through the second communication interface.
8. The method according to claim 6, wherein the routing function module sends the first data packet to the first neighboring artificial intelligence chip through the first communication interface, specifically comprising:
the routing function module sends the first data packet and the first transmission path to the first adjacent artificial intelligent chip through the first communication interface, so that the first adjacent artificial intelligent chip executes the transmission operation of the first data packet based on the first transmission path.
9. A data transmission system comprising a plurality of artificial intelligence chips; the data transmission system is constructed based on a cross switch matrix mode; the artificial intelligent chips are interconnected in a cross switch matrix mode; the artificial intelligence chip is used for executing the data transmission method as claimed in claim 4 or 5.
10. A data transmission system comprising a plurality of artificial intelligence chips; the data transmission system is constructed based on a wireless network grid mode; the artificial intelligent chips are interconnected in a two-dimensional wireless network grid mode or in a three-dimensional wireless network grid mode; the artificial intelligence chip is for performing the data transmission method of any one of claims 6-8.
CN202310877938.0A 2023-07-18 2023-07-18 Artificial intelligent chip, data transmission method and data transmission system Active CN116614433B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310877938.0A CN116614433B (en) 2023-07-18 2023-07-18 Artificial intelligent chip, data transmission method and data transmission system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310877938.0A CN116614433B (en) 2023-07-18 2023-07-18 Artificial intelligent chip, data transmission method and data transmission system

Publications (2)

Publication Number Publication Date
CN116614433A true CN116614433A (en) 2023-08-18
CN116614433B CN116614433B (en) 2023-10-24

Family

ID=87675055

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310877938.0A Active CN116614433B (en) 2023-07-18 2023-07-18 Artificial intelligent chip, data transmission method and data transmission system

Country Status (1)

Country Link
CN (1) CN116614433B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104052663A (en) * 2013-03-14 2014-09-17 中国人民解放军信息工程大学 Large-scale on-chip chip interconnecting method and routing algorithm for realizing interconnecting structure
CN111177065A (en) * 2018-11-12 2020-05-19 深圳市中兴微电子技术有限公司 Multi-chip interconnection method and device
US20220083498A1 (en) * 2018-12-28 2022-03-17 Lynxi Technologies Co., Ltd. Data transmission method and device for network on chip and electronic apparatus
CN115115043A (en) * 2022-06-20 2022-09-27 上海交通大学 Method and system for designing hardware architecture of on-chip-to-chip interconnection neural network chip

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104052663A (en) * 2013-03-14 2014-09-17 中国人民解放军信息工程大学 Large-scale on-chip chip interconnecting method and routing algorithm for realizing interconnecting structure
CN111177065A (en) * 2018-11-12 2020-05-19 深圳市中兴微电子技术有限公司 Multi-chip interconnection method and device
US20220083498A1 (en) * 2018-12-28 2022-03-17 Lynxi Technologies Co., Ltd. Data transmission method and device for network on chip and electronic apparatus
CN115115043A (en) * 2022-06-20 2022-09-27 上海交通大学 Method and system for designing hardware architecture of on-chip-to-chip interconnection neural network chip

Also Published As

Publication number Publication date
CN116614433B (en) 2023-10-24

Similar Documents

Publication Publication Date Title
CN113098773B (en) Data processing method, device and system
CN109451540B (en) Resource allocation method and equipment for network slices
US8166202B2 (en) Computer supporting remote scan
CN110138596A (en) A kind of block chain common recognition method based on handover network topology mode
US20220210030A1 (en) SYNTHESIS OF A NETWORK-ON-CHIP (NoC) USING PERFORMANCE CONSTRAINTS AND OBJECTIVES
CN102685017A (en) On-chip network router based on field programmable gate array (FPGA)
CN101808032A (en) Static XY routing algorithm-oriented two-dimensional grid NoC router optimization design method
CN107707628B (en) Method and apparatus for transmitting data processing requests
CN102780628A (en) On-chip interconnection network routing method oriented to multi-core microprocessor
CN105530206B (en) A kind of Torus network system and its working method with double access infrastructures
US20180181536A1 (en) Cpu interconnect apparatus and system, and cpu interconnect control method and control apparatus
CN114707451A (en) Digital circuit layout planning method and device, electronic equipment and storage medium
CN117061423B (en) Multi-machine routing method, device and system of fat tree network and storage medium
CN116614433B (en) Artificial intelligent chip, data transmission method and data transmission system
CN116383114B (en) Chip, chip interconnection system, data transmission method, electronic device and medium
CN103299298A (en) Service processing method and system
CN116915708A (en) Method for routing data packets, processor and readable storage medium
US9762474B2 (en) Systems and methods for selecting a router to connect a bridge in the network on chip (NoC)
CN113691403B (en) Topology node configuration method, related device and computer program product
CN110825689B (en) Electronic chip and implementation method thereof
CN112463670A (en) Storage controller access method and related device
CN115955429B (en) Network-on-chip routing method, device, system and electronic equipment
CN116306407B (en) Verification method, device, equipment and storage medium of Network On Chip (NOC)
CN115665041B (en) Network-on-chip structure, data transmission method, electronic device, and storage medium
CN114095289B (en) Data multicast circuit, method, electronic device, and computer-readable storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant