On-chip network, chip, and data transmission method in chip

By designing multiple subnets in the on-chip network and configuring different data forwarding rules, the poor connectivity and deadlock of functional modules during regional failure of the chip are solved, and data transmission with low latency and high connectivity is achieved.

CN117951079BActive Publication Date: 2024-11-19NEW H3C SEMICON TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410348957.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-26
Publication Date
2024-11-19
Estimated Expiration
2044-03-26

AI Technical Summary

Technical Problem

The existing on-chip network ignores the connectivity between functional modules when the chip regional failure is designed, resulting in high communication delays and prone to deadlocks.

Method used

Design an on-chip network containing at least two on-chip subnets. Each subnet is pre-configured with different data forwarding rules. The routing module forwards data according to the rules of the subnet to ensure that the sub-module passes different rules when the functional module is blocked. The network implements data transmission, avoids deadlocks and maintains low latency and high connectivity.

Benefits of technology

When the functional modules experience regional failure, data transmission is achieved through on-chip sub-networks with different data forwarding rules to ensure low latency and high connectivity between the functional modules and avoid deadlock.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117951079B_ABST
    Figure CN117951079B_ABST
Patent Text Reader

Abstract

The embodiments of the present application provide an on-chip network, a chip, and a data transmission method in a chip, wherein the on-chip network includes at least two on-chip sub-networks, and the on-chip sub-networks include a routing module; different data forwarding rules are pre-configured for different on-chip sub-networks; the routing module is used to forward data in accordance with the data forwarding rules of the on-chip sub-network to which it belongs; the data comes from the functional modules connected to the chip itself, or from the adjacent routing modules belonging to the same on-chip sub-network. When a regional failure occurs in the functional modules of the chip, for the data that needs to be sent from the source functional module to the destination functional module, if one of the on-chip sub-networks cannot realize data forwarding by following the data forwarding rules, data forwarding can be realized by an on-chip sub-network configured with different data forwarding rules. Thus, when some functional modules are shielded, low latency and high connectivity for access between functional modules are still guaranteed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to an on-chip network, a chip, and a data transmission method in a chip. Background Art

[0002] On-Chip Networks (NoC) are a major component of multi-core technology and are gradually becoming a ubiquitous communication structure in multi-core chips. On-Chip Networks consist of transmission channels and router nodes. The network topology determines the physical layout and connections between nodes and channels in the network.

[0003] When designing an on-chip network, the following basic requirements need to be considered: 1) Minimize the communication latency between cores. 2) Ensure that data transmission between cores does not deadlock as much as possible. While meeting the above basic requirements, current on-chip networks often ignore the interconnectivity between cores when a chip fails locally. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide an on-chip network, a chip, and a data transmission method in a chip, which is used to improve the connectivity between functional modules when regional timeliness occurs in the chip while minimizing the communication delay between functional modules and ensuring that data transmission is not deadlocked.

[0005] In a first aspect of the present application, a network on chip is provided.

[0006] The on-chip network includes at least two on-chip sub-networks; the on-chip sub-network includes a routing module; different data forwarding rules are pre-configured for different on-chip sub-networks;

[0007] The routing module is used to forward data in accordance with the data forwarding rules of the on-chip sub-network to which it belongs; the data comes from the functional modules connected to the chip itself, or from the adjacent routing modules belonging to the same on-chip sub-network.

[0008] Optionally, the on-chip sub-network is a mesh topology.

[0009] Optionally, the data forwarding rules are used to limit the direction of data forwarding by the routing module so that the data transmission of the on-chip sub-network is deadlock-free; the direction includes: from a first direction to a second direction, and from the second direction to the first direction; wherein, the first direction is the horizontal direction of the mesh topology structure; and the second direction is the vertical direction of the mesh topology structure.

[0010] Optionally, the on-chip network includes a first on-chip sub-network and a second on-chip sub-network.

[0011] Optionally, for the first routing module and the second routing module connected to the same functional module, the data forwarding rule of the first on-chip sub-network restricts the redirection of the first routing module, which is different from the data forwarding rule of the second on-chip sub-network that restricts the redirection of the second routing module; wherein, the first routing module belongs to the first on-chip sub-network, and the second routing module belongs to the second on-chip sub-network.

[0012] Optionally, data forwarding rules configured for different on-chip sub-networks meet the following conditions:

[0013] For any first data sent from a source functional module to a destination functional module, when there is a blocked functional module, there is at least one target on-chip sub-network that follows the configured data forwarding rules to complete the forwarding of the first data, and the forwarding path does not pass through the blocked functional module.

[0014] Optionally, when the functional module is used as a source functional module for sending data, it is used to determine the target on-chip sub-network according to the routing table; there is a target path that follows the data forwarding rule in the target on-chip sub-network; the target path is the path from the source functional module to the destination functional module; the routing table maintains the reachable paths that follow the data forwarding rule in each on-chip sub-network;

[0015] The functional module is also used to send data to the routing module connected to the routing module and belonging to the target on-chip sub-network; the data carries a forwarding instruction generated based on the target path;

[0016] The routing module is specifically used to forward data according to the forwarding instruction.

[0017] Optionally, when there is a blocked functional module in the chip; the routing table maintains reachable paths in each on-chip sub-network that follow the data forwarding rule and do not pass through the blocked functional module;

[0018] The source functional module is specifically used to determine the target on-chip sub-network according to the reachable paths of the functional modules that follow the data forwarding rules and are not blocked in each on-chip sub-network maintained in the routing table, wherein there is a target path of the functional modules that follow the data forwarding rules and are not blocked in the target on-chip sub-network; the target path is the path from the source functional module to the destination functional module.

[0019] A second aspect of the present application provides a chip, including a functional module and any of the above-mentioned on-chip networks.

[0020] In a third aspect of the present application, a data transmission method in a chip is provided.

[0021] The chip includes: a functional module, an on-chip network, and the on-chip network includes a routing module; different data forwarding rules are pre-configured for different on-chip sub-networks;

[0022] The method comprises:

[0023] The source functional module determines the target on-chip sub-network according to the routing table; the target on-chip sub-network has a target path that follows the data forwarding rule; the target path is the path from the source functional module to the destination functional module; the routing table maintains the reachable paths that follow the data forwarding rule in each on-chip sub-network;

[0024] The source functional module sends data to the routing module connected to itself and belonging to the target on-chip sub-network; the data carries forwarding instructions generated based on the target path, so that the routing module of the target on-chip sub-network sends the data to the destination functional module according to the forwarding instructions.

[0025] Optionally, when there is a blocked functional module in the chip;

[0026] The routing table maintains reachable paths in each on-chip sub-network that follow data forwarding rules and do not pass through blocked functional modules;

[0027] The step of the source function module determining the target on-chip sub-network according to the routing table comprises:

[0028] The source functional module determines the target on-chip sub-network according to the reachable paths of the functional modules that follow the data forwarding rules and are not blocked in each on-chip sub-network maintained in the routing table, wherein there is a target path in the target on-chip sub-network that follows the data forwarding rules and is not blocked by the functional modules; the target path is the path from the source functional module to the destination functional module.

[0029] The on-chip network provided in the embodiment of the present application includes at least two on-chip sub-networks, and different data forwarding rules are configured for different on-chip sub-networks. The routing module forwards data in accordance with the data forwarding rules of the on-chip sub-network to which it belongs. When a regional failure occurs in a functional module of the chip, for data that needs to be sent from the source functional module to the destination functional module, if one of the on-chip sub-networks cannot forward data in accordance with the data forwarding rules, data forwarding can be achieved through an on-chip sub-network configured with different data forwarding rules. Therefore, when some functional modules are shielded, low latency and high connectivity for access between functional modules are still guaranteed.

[0030] Of course, it is not necessary to achieve all of the advantages described above at the same time to implement any product or method of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other embodiments can also be obtained based on these drawings.

[0032] Figure 1 A schematic diagram of a through-type chip structure;

[0033] Figure 2 A schematic diagram of a store-and-forward chip structure;

[0034] Figure 3 A schematic diagram of a structure of an on-chip network in the related art;

[0035] Figure 4 is another structural schematic diagram of an on-chip network in the related art;

[0036] Figure 5 A schematic diagram of data deadlock in on-chip network;

[0037] Figure 6 A schematic diagram of the structure of a network on chip provided in an embodiment of the present application;

[0038] Figure 7 A schematic diagram of data transmission when a functional module in an on-chip network is blocked;

[0039] Figure 8 Another schematic diagram of data transmission when a functional module in an on-chip network is blocked;

[0040] Fig. 9 A flowchart of a data transmission method in a chip provided in an embodiment of the present application. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field based on the present invention belong to the scope of protection of the present invention.

[0042] For ease of understanding, the relevant technologies are introduced first.

[0043] For the core network processing chip in network forwarding equipment, there are usually two data processing solutions:

[0044] 1) Direct-through mode. Direct-through mode is to directly transmit message data between each internal module, which is simple to implement. Figure 1 As shown, Figure 1 This is a schematic diagram of a direct chip structure, showing module 1, module 2 and module 3. However, the internal data bandwidth requirement is high, and there is bandwidth waste for processing flows that only require part of the message information. For example, module 2 only requires part of the message data, but obtains all the message data; at the same time, since there is no external storage such as DRAM (Dynamic Random Access Memory), the internal storage capacity of the chip is limited by the chip area. This mode of processing chip is difficult to meet the business processing that requires a large cache capacity.

[0045] 2) Store-and-forward. The store-and-forward method uses external storage such as DDR SDRAM (Double Data Rate Synchronous Dynamic Random Access Memory) and HBM (High Bandwidth Memory). The message is first stored in the external storage, and the edge module generates simplified description information for transmission. See Figure 2 , Figure 2 A schematic diagram of a store-and-forward chip structure is shown in FIG. Figure 2 As shown in the figure, if the message data is 200 bytes in total, module 1 only transmits a small part of the data to module 2, such as 10 bytes. This extracted information is sufficient to complete the data processing of most modules. The line width between modules only needs to meet the descriptor width. The chip area saved by this solution can be used for more complex processing flows. When the processing is completed, the message is extracted from the external storage and sent.

[0046] Gradually, some modules in the network chip have implemented universal design and introduced software instructions to achieve business flexibility. Some processing modules in the middle have been converted from dedicated hardware design to universal software core.

[0047] The more complex the processing flow, the longer it takes to process a single message. Complex router services also introduce a large number of software operations, which requires maximizing the chip's processing parallelism to meet high-bandwidth forwarding performance. Bandwidth = message length / processing time × processing parallelism. How to organize the interconnection between a large number of processing cores in a network chip, as well as the connection between the processing core and external storage, so that access conflicts can be avoided as much as possible while maintaining a reasonable number of connections, has become a difficult and key point in the design of network processing chips.

[0048] See also Figure 3 , which shows a schematic diagram of a structure of an on-chip network in the related art, such as Figure 3As shown in the figure, a bus structure is created between the processing core and the peripheral modules. The access of message data and the interaction between modules and cores, and between cores are all carried out through the bus. The bus has a natural broadcast attribute, so it is possible to achieve multiple points of reception of one data without copying the data. In terms of implementation, the bus consists of two unidirectional data paths in opposite directions to ensure that data transmission will not conflict or deadlock. The transmission destination is determined by addressing the bus access point, and the sender calculates and determines the transmission direction with the shorter path.

[0049] However Figure 3 The on-chip network structure shown is not easy to expand, and as the number of processing cores increases, the number of routing points the bus passes through will also increase, resulting in a higher latency for a single data transmission.

[0050] See also Figure 4 ,Another structural diagram of the on-chip network in the related art, such as Figure 4 As shown in the figure, mesh routing is created between the processing cores, where mesh refers to a grid-like topology. The regularly arranged routing points are connected to each other through a bidirectional transmission channel, and each routing point can perform three operations: turning left, turning right, and going straight. The routing points are marked with coordinates, and the sender creates a series of turn and straight instructions and carries them in the data. Each time a routing point is passed, an instruction information is removed until the transmission destination is reached.

[0051] However Figure 4 The on-chip network structure shown in the figure has the risk of path deadlock. Figure 5 , is a schematic diagram of path deadlock in mesh structure, such as Figure 5 As shown, at the same time, the data sent from routing module S1 to routing module D1 arrives at routing module a, the data sent from routing module S2 to routing module D2 arrives at routing module b, the data sent from routing module S3 to routing module D3 arrives at routing module c, and the data sent from routing module S4 to routing module D4 arrives at routing module d. Each group of data needs to wait for the next group of data to be forwarded, resulting in a circular wait and causing a path deadlock.

[0052] In addition, since the functional modules in the chip may be blocked, it is also crucial to ensure the interconnection between other functional modules when the functional modules are blocked. The current on-chip network often ignores the interconnection between functional modules when the chip fails regionally.

[0053] In order to solve the above technical problems, the embodiments of the present application provide an on-chip network, a chip, and a data transmission method in a chip.

[0054] Among them, the on-chip network is also called the on-chip bus. The on-chip network includes transmission channels and routing modules. The routing module can also be understood as a router node (router), which is used to connect various functional modules in the chip. The functional modules in the chip are modules that combine software and hardware to implement specific functions, such as CPU core, packet decoder, memory controller, queue manager, etc.

[0055] The on-chip network provided in the embodiment of the present application includes at least two on-chip sub-networks; the on-chip sub-network includes a routing module; different data forwarding rules are pre-configured for different on-chip sub-networks; the routing module is used to forward data in accordance with the data forwarding rules of the on-chip sub-network to which it belongs; the data comes from the functional modules connected to the chip itself, or from the adjacent routing modules belonging to the same on-chip sub-network.

[0056] As an example, see Figure 6 , Figure 6 A schematic diagram of a structure of a network on chip provided in an embodiment of the present application, Figure 6 A functional module 601 and two on-chip sub-networks are shown.

[0057] Each on-chip sub-network includes a routing module 602 and a transmission channel for connecting adjacent routing modules 602 .

[0058] Figure 6 In the example shown, in order to more intuitively distinguish the first on-chip sub-network from the second on-chip sub-network, a solid line and a dotted line are used to represent the transmission channel of the first on-chip sub-network and the transmission channel of the second on-chip sub-network, respectively. Figure 6 As shown, the first on-chip sub-network includes 9 routing modules interconnected by solid lines, and the second on-chip sub-network includes 9 routing modules interconnected by dotted lines.

[0059] In addition, the routing module 602 is connected to the specific functional module 601 via a transmission channel to achieve data transmission between the routing module 602 and the functional module 601 .

[0060] In an embodiment of the present application, the routing module is used to forward data in accordance with the data forwarding rules of the on-chip sub-network to which it belongs; the data comes from the functional modules connected to the chip itself, or from the adjacent routing modules belonging to the same on-chip sub-network.

[0061] In the embodiment of the present application, each functional module can be connected to at least two routing modules belonging to different on-chip sub-networks.

[0062] Figure 6Nine functional modules 601 are shown, wherein each functional module 601 is connected to a routing module belonging to the first on-chip sub-network and the second on-chip sub-network respectively.

[0063] In the embodiment of the present application, different data forwarding rules are pre-configured for different on-chip sub-networks. Figure 5 In the deadlock situation shown, data forwarding rules can be configured in advance for each on-chip sub-network to ensure that data deadlock will not occur in each on-chip sub-network.

[0064] That is, the data forwarding rules configured for the on-chip sub-network are used to ensure that no data deadlock occurs when the routing module of the on-chip sub-network follows the data forwarding rules to forward data. The embodiments of the present application do not limit the specific data forwarding rules.

[0065] Since data forwarding rules are configured to prevent data deadlock, when a functional module in the chip is blocked, the connectivity between other functional modules in the chip may be greatly affected. It should be noted that a functional module is blocked, which means that the functional module has a regional failure, that is, the data channels in the area where the functional module is located cannot be connected, including the data channels in the on-chip bus that pass through the area.

[0066] For ease of understanding, the following Figure 7 Provide explanation. Figure 7 A schematic diagram of data transmission when a functional module in an on-chip network is blocked. Figure 7 9 routing modules and the data channels between the routing modules are shown in FIG. 1 . To prevent data deadlock, assume that the configured data forwarding rules include: the routing modules in odd columns only allow horizontal to vertical diversion, and the routing modules in even columns only allow vertical to horizontal diversion. Then in Figure 7 When the functional modules in the dotted box are blocked, if routing module A needs to transmit data to routing module D, it can only pass the data to routing module B first. Since routing module B is in an even column, horizontal to vertical steering is not allowed, so routing module B cannot transmit the data to routing module C. If routing module B transmits data to routing module E, routing module E can further transmit the data to routing module F or routing module G, but routing module F and routing module G are both in odd columns, vertical to horizontal steering is not allowed, so data cannot be transmitted to routing module C and routing module D.

[0067] It can be seen that when the functional modules in the dotted box are blocked, routing module A and routing module C cannot be connected, and routing module A and routing module D cannot be connected, which causes a significant impact on the communication between the functional modules.

[0068] In the embodiment of the present application, different data forwarding rules are pre-configured for different on-chip sub-networks. Then, when a functional module is blocked, for data that needs to be sent from the source functional module to the destination functional module, if one of the on-chip sub-networks cannot forward the data according to the data forwarding rules, the data can be forwarded through the on-chip sub-network configured with different data forwarding rules.

[0069] For ease of understanding, the following Figure 7 and attached Figure 8 Provide explanation. Figure 8 Another schematic diagram of data transmission when functional modules in the on-chip network are blocked. Figure 7 and attached Figure 8 Each shows an on-chip sub-network and an on-chip network belonging to the same chip. Figure 8 The nine routing modules shown are Figure 7 The 9 routing modules in the table correspond one to one. Figure 8 Routing module a, attached Figure 7 The routing module A in the same functional module is connected. Figure 8 Routing module b in the attachment Figure 7 The routing module B in connects to the same functional module, and so on.

[0070] For attachment Figure 8 In order to prevent data deadlock, the configured data forwarding rules for the on-chip sub-network shown in the figure include: the routing modules in odd-numbered columns only allow vertical to horizontal switching, and the routing modules in even-numbered columns only allow horizontal to vertical switching. Figure 8 When the functional module in the dotted box is blocked, if routing module a needs to transmit data to routing module d, it can only pass the data to routing module b first. Since routing module b is in an even column, horizontal to vertical steering is allowed, so routing module b can transmit the data to routing module c, and then routing module c transmits the data to routing module d.

[0071] It can be seen that when the functional modules in the dotted box are blocked, routing module a and routing module c can be connected, and routing module a and routing module d can also be connected.

[0072] The on-chip network provided in the embodiment of the present application includes at least two on-chip sub-networks, and different data forwarding rules are configured for different on-chip sub-networks. The routing module forwards data in accordance with the data forwarding rules of the on-chip sub-network to which it belongs. When a regional failure occurs in a functional module of the chip, for data that needs to be sent from the source functional module to the destination functional module, if one of the on-chip sub-networks cannot forward data in accordance with the data forwarding rules, data forwarding can be achieved through an on-chip sub-network configured with different data forwarding rules. Therefore, when some functional modules are shielded, low latency and high connectivity for access between functional modules are still guaranteed.

[0073] In one embodiment of the present application, the on-chip sub-network is a mesh topology.

[0074] Specifically, the grid topology is a mesh structure. The grid topology is easy to expand, and the access latency will not increase exponentially with the expansion of the core scale.

[0075] In one embodiment of the present application, data forwarding rules are used to limit the direction of data forwarding by the routing module so that data transmission in the on-chip sub-network is deadlock-free; the direction includes: from a first direction to a second direction, and from the second direction to the first direction; wherein the first direction is the horizontal direction of the mesh topology structure; and the second direction is the vertical direction of the mesh topology structure.

[0076] Specifically, the on-chip sub-network is a mesh topology structure, so there are horizontal and vertical directions. When configuring data forwarding rules, the direction of data forwarding by the routing module can be restricted.

[0077] Since the grid topology structure has horizontal and vertical directions, there are two types of data rotation: from horizontal to vertical, and from vertical to horizontal.

[0078] When configuring data forwarding rules, by limiting the direction of data forwarding by the routing module, it is possible to ensure that data transmission in a single on-chip sub-network is deadlock-free.

[0079] As an example, for a single on-chip sub-network, the data transmission of the routing module is restricted to only support the transition from the horizontal direction to the vertical direction, so that the on-chip sub-network will not have a data transmission deadlock.

[0080] It can be seen that in the embodiment of the present application, when the on-chip sub-network is a mesh topology structure, the direction of forwarding data by the routing module can be restricted to make the data transmission of the on-chip sub-network deadlock-free and avoid the scenario of circular deadlock waiting.

[0081] In one embodiment of the present application, the on-chip network includes a first on-chip sub-network and a second on-chip sub-network. When the on-chip network includes the first on-chip sub-network and the second on-chip sub-network, for a first routing module and a second routing module connected to the same functional module, the data forwarding rule of the first on-chip sub-network restricts the redirection of the first routing module, which is different from the data forwarding rule of the second on-chip sub-network restricting the redirection of the second routing module; wherein the first routing module belongs to the first on-chip sub-network, and the second routing module belongs to the second on-chip sub-network.

[0082] As an example, the data forwarding rules of the first on-chip sub-network restrict the routing module to only support forwarding data with a first redirection, and the data forwarding rules of the second on-chip sub-network restrict the routing module to only support forwarding data with a second redirection, wherein the first redirection is different from the second redirection.

[0083] For example, the routing module of the first on-chip sub-network only supports the transition from the horizontal direction to the vertical direction, so no data deadlock will occur inside the first on-chip sub-network; the routing module of the second on-chip sub-network only supports the transition from the vertical direction to the horizontal direction, so no data deadlock will occur inside the second on-chip sub-network.

[0084] As another example, the data forwarding rules of the first on-chip sub-network restrict the routing modules of odd-numbered columns to only support forwarding data with the first turn, and the routing modules of even-numbered columns to only support forwarding data with the second turn; the data forwarding rules of the second on-chip sub-network restrict the routing modules of odd-numbered columns to only support forwarding data with the second turn, and the routing modules of even-numbered columns to only support forwarding data with the first turn.

[0085] Similarly, data deadlock will not occur in the first on-chip sub-network and the second on-chip sub-network.

[0086] Furthermore, since the data forwarding rules of the first on-chip sub-network and the second on-chip sub-network are different, the data forwarding path of the first on-chip sub-network and the data forwarding path of the second on-chip sub-network can complement each other. If one of the on-chip sub-networks cannot achieve data forwarding by following the data forwarding rules, data forwarding can be achieved through an on-chip sub-network configured with different data forwarding rules. Thus, when some functional modules are shielded, low latency and high connectivity for access between functional modules are still guaranteed.

[0087] In one embodiment of the present application, data forwarding rules configured for different on-chip sub-networks meet the following conditions:

[0088] For any first data sent from a source functional module to a destination functional module, when there is a blocked functional module, there is at least one target on-chip sub-network that follows the configured data forwarding rules to complete the forwarding of the first data, and the forwarding path does not pass through the blocked functional module.

[0089] In a possible embodiment, the blocked functional modules in the chip are determined at the factory. Therefore, the location of the blocked functional modules can be considered when configuring data forwarding rules. For any first data sent from the source functional module to the destination functional module, it is guaranteed that there is at least one target on-chip sub-network, which follows the configured data forwarding rules to complete the forwarding of the first data, and the forwarding path does not pass through the blocked functional module.

[0090] In addition, it is worth noting that if it is to be ensured that any first data sent from the source functional module to the destination functional module is transmitted through only one target on-chip sub-network, it may be necessary to build more on-chip sub-networks, which increases the complexity and cost of the on-chip network to a certain extent.

[0091] Therefore, when designing on-chip sub-networks and data forwarding rules, it is sufficient to ensure the connectivity of most routing nodes in a single on-chip sub-network. For data to be transmitted with a long path between the source functional module and the destination functional module, it can be transmitted through multiple on-chip sub-networks. See below for details.

[0092] In one embodiment of the present application, when the functional module acts as a source functional module for sending data, it is used to determine the target on-chip sub-network according to the routing table; there is a target path that follows the data forwarding rules in the target on-chip sub-network; the target path is the path from the source functional module to the destination functional module; the routing table maintains the reachable paths that follow the data forwarding rules in each on-chip sub-network.

[0093] Specifically, since the topology and data forwarding rules of each on-chip sub-network are determined, the reachable paths that follow the data forwarding rules in each on-chip sub-network are also determined, and the reachable paths that follow the data forwarding rules in each on-chip sub-network can be maintained in the routing table.

[0094] The routing table may be stored in each functional module, or a storage module for storing the routing table may be provided in the chip, and the functional module accesses the storage module when needed to read the information in the routing table.

[0095] When the functional module acts as the source functional module for sending data, the source functional module determines the target on-chip sub-network according to the information maintained in the routing table, that is, under the premise of following the data forwarding rules of the target on-chip sub-network, the data sent by the source functional module can be forwarded to the destination functional module.

[0096] There may be more than one target on-chip subnetwork to be determined, and the one with the smallest hop count may be selected first, or load sharing of multiple on-chip subnetworks may be considered, that is, the data to be forwarded may be distributed to different on-chip subnetworks as much as possible.

[0097] After determining the target on-chip sub-network and the target path, the source functional module generates a forwarding instruction based on the target path. The forwarding instruction may include continuous "turn left", "turn right", and "go straight" operation instructions. The data containing the forwarding instruction is sent to the routing module of the target on-chip sub-network connected to itself, and the routing module forwards it according to the forwarding instruction. For example, the routing module forwards the data according to the operation instruction, and removes an operation instruction each time it passes through a routing module until the data is forwarded to the routing module connected to the destination functional module, and the routing module then forwards the data to the destination functional module.

[0098] It can be seen that in the embodiment of the present application, the on-chip network includes multiple on-chip sub-networks, and different data forwarding rules are configured for each on-chip sub-network, so that the reachable paths that follow the data forwarding rules of each on-chip sub-network are also different. Therefore, when the source functional module needs to forward data to the destination functional module, it can select the on-chip sub-network that can complete the data forwarding according to the reachable paths of each on-chip sub-network maintained in the routing table. It can be seen that the reachable paths of different on-chip sub-networks complement each other, and under the premise that there is no deadlock in each on-chip sub-network, low latency and high connectivity between functional modules are guaranteed.

[0099] In one embodiment of the present application, when there is a blocked functional module in the chip, the routing table maintains reachable paths in each on-chip sub-network that follow data forwarding rules and do not pass through the blocked functional module.

[0100] Specifically, when there is a blocked functional module in the chip, the data channel contained in the on-chip sub-network cannot pass through the area where the blocked functional module is located, resulting in a reduction in reachable paths. In this case, the routing table maintains reachable paths in each on-chip sub-network that follow the data forwarding rules and do not pass through the blocked functional modules. Only these reachable paths can forward data.

[0101] In a possible embodiment, the blocked functional modules in the chip are determined at the factory. In this case, maintaining the reachable paths in the routing table may be pre-configured.

[0102] Accordingly, the source functional module determines the target on-chip sub-network according to the reachable paths of the functional modules that follow the data forwarding rules and do not pass through the blocked functional modules in each on-chip sub-network maintained in the routing table. That is, under the premise of following the data forwarding rules of the target on-chip sub-network and not passing through the area where the blocked functional modules are located, the data sent by the source functional module can be forwarded to the destination functional module.

[0103] It can be seen that in the embodiment of the present application, the on-chip network includes multiple on-chip sub-networks. For each on-chip sub-network, the configured data forwarding rules are different, so that the reachable paths that follow the data forwarding rules of each on-chip sub-network are also different. If one of the on-chip sub-networks cannot achieve data forwarding by following the data forwarding rules, data forwarding can be achieved through an on-chip sub-network configured with different data forwarding rules. And when some functional modules are shielded, the reachable paths of the functional modules that follow the data forwarding rules and are not blocked can be selected according to the routing table, thereby ensuring low latency and high connectivity between functional modules.

[0104] In one embodiment of the present application, the functional module is further used to: forward data from the routing module connected to itself and belonging to the first on-chip sub-network to the routing module connected to itself and belonging to the second on-chip sub-network.

[0105] Specifically, for data sent from a source functional module to a destination functional module, if data transmission cannot be completed through only a single on-chip sub-network, the data is allowed to be transmitted through more than one on-chip sub-network, that is, the functional module can serve as a relay node for data transmission, forwarding data from one on-chip sub-network to data from another on-chip sub-network.

[0106] As an example, source functional module A needs to send data to destination functional module B. According to the information maintained in the routing table, any reachable path of the on-chip sub-network cannot complete the data transmission. However, with the help of the reachable path of the first on-chip sub-network, the data can be sent from the source functional module A to the functional module C. With the help of the reachable path of the second on-chip sub-network, the data can be sent from the functional module C to the destination functional module B. Then, functional module C can act as a relay node to forward the data from the first on-chip sub-network to the second on-chip sub-network, thereby forwarding the data through more than one on-chip sub-network.

[0107] It can be seen that in the embodiments of the present application, the functional module can act as a relay node to forward data from one on-chip sub-network to another on-chip sub-network, thereby transmitting data through more than one on-chip sub-network, which can further improve the high connectivity between the functional modules without having to configure too many on-chip sub-networks, thereby reducing the complexity and cost of the on-chip network.

[0108] The embodiment of the present application further provides a chip, including a functional module and an on-chip network, wherein the on-chip network includes at least two on-chip sub-networks; the on-chip sub-network includes a routing module; different data forwarding rules are pre-configured for different on-chip sub-networks;

[0109] The routing module is used to forward data in accordance with the data forwarding rules of the on-chip sub-network to which it belongs; the data comes from the functional modules connected to itself in the chip, or from the adjacent routing modules belonging to the same on-chip sub-network.

[0110] When a chip's functional module fails in a certain area, if one of the on-chip sub-networks cannot forward data according to the data forwarding rules, data forwarding can be achieved through an on-chip sub-network configured with different data forwarding rules. This ensures low latency and high connectivity between functional modules even when some functional modules are shielded.

[0111] See also Fig. 9 , is a flow chart of a data transmission method in a chip provided in an embodiment of the present application, comprising the following steps:

[0112] S901: The source functional module determines the target on-chip sub-network according to the routing table; the target on-chip sub-network has a target path that follows the data forwarding rules; the target path is the path from the source functional module to the destination functional module; the routing table maintains the reachable paths that follow the data forwarding rules in each on-chip sub-network.

[0113] S902: The source functional module sends data to the routing module connected to itself and belonging to the target on-chip sub-network; the data carries a forwarding instruction generated based on the target path, so that the routing module of the target on-chip sub-network sends the data to the destination functional module according to the forwarding instruction.

[0114] In the embodiment of the present application, since the on-chip network includes multiple on-chip sub-networks, the data forwarding rules configured for each on-chip sub-network are different, so that the reachable paths that follow the data forwarding rules of each on-chip sub-network are also different. Therefore, when the source functional module needs to forward data to the destination functional module, it can select the on-chip sub-network that can complete the data forwarding according to the reachable paths of each on-chip sub-network maintained in the routing table. It can be seen that the reachable paths of different on-chip sub-networks complement each other, and under the premise that there is no deadlock in each on-chip sub-network, low latency and high connectivity between functional modules are guaranteed.

[0115] In one embodiment of the present application, when there is a blocked functional module in the chip, the routing table maintains reachable paths in each on-chip sub-network that follow data forwarding rules and do not pass through the blocked functional module.

[0116] Accordingly, the source functional module determines the target on-chip sub-network according to the routing table, including:

[0117] The source functional module determines the target on-chip sub-network according to the reachable paths of the functional modules that follow the data forwarding rules and do not pass through blocking in each on-chip sub-network maintained in the routing table, wherein the target on-chip sub-network has a target path that follows the data forwarding rules and does not pass through blocking functional modules; the target path is the path from the source functional module to the destination functional module.

[0118] Specifically, when there is a blocked functional module in the chip, the routing table may maintain reachable paths in each on-chip sub-network that follow the data forwarding rule and do not pass through the blocked functional module.

[0119] Therefore, when determining the target on-chip sub-network, it is determined that there is an on-chip sub-network with a target path that follows the data forwarding rules and does not pass through the blocked functional modules. That is to say, transmitting data through the target on-chip sub-network can avoid the blocked functional modules and realize normal data transmission between the source functional modules and the destination functional modules.

[0120] It can be seen that when a functional module of the chip fails in a regional manner, for data that needs to be sent from the source functional module to the destination functional module, if one of the on-chip sub-networks follows the data forwarding rules and cannot forward the data, data forwarding can be achieved through an on-chip sub-network configured with different data forwarding rules. In this way, when some functional modules are shielded, low latency and high connectivity for access between functional modules are still guaranteed.

[0121] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0122] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0123] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, electronic device, and storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0124] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A network on chip, characterized in that: The on-chip network includes a functional module and at least two on-chip sub-networks; the on-chip sub-network includes a routing module; different data forwarding rules are pre-configured for different on-chip sub-networks; each of the functional modules is connected to the routing modules belonging to the first on-chip sub-network and the second on-chip sub-network respectively; The routing module is used to forward data in accordance with the data forwarding rules of the on-chip sub-network to which it belongs; the data comes from the functional modules connected to the chip itself, or from the adjacent routing modules belonging to the same on-chip sub-network; The on-chip sub-network is a mesh topology structure; The data forwarding rule is used to limit the direction of data forwarding by the routing module so that the data transmission of the on-chip sub-network is deadlock-free; the direction includes: from a first direction to a second direction, and from a second direction to a first direction; wherein the first direction is the horizontal direction of the mesh topology structure; and the second direction is the vertical direction of the mesh topology structure; The on-chip network includes a first on-chip sub-network and a second on-chip sub-network; For a first routing module and a second routing module connected to the same functional module, a data forwarding rule of the first on-chip sub-network restricts the redirection of the first routing module, which is different from a data forwarding rule of the second on-chip sub-network restricting the redirection of the second routing module; wherein the first routing module belongs to the first on-chip sub-network, and the second routing module belongs to the second on-chip sub-network; When the functional module is used as a source functional module for sending data, it is used to determine the target on-chip sub-network according to the routing table; there is a target path that follows the data forwarding rule in the target on-chip sub-network; the target path is the path from the source functional module to the destination functional module; the routing table maintains the reachable paths that follow the data forwarding rule in each on-chip sub-network; The functional module is also used to send data to the routing module connected to the routing module and belonging to the target on-chip sub-network; the data carries a forwarding instruction generated based on the target path; The routing module is specifically used to forward data according to the forwarding instruction; The functional module is also used to: forward data from the routing module connected to itself and belonging to the first on-chip sub-network to the routing module connected to itself and belonging to the second on-chip sub-network; When there is a blocked functional module in the chip; the routing table maintains reachable paths in each on-chip sub-network that follow the data forwarding rules and do not pass through the blocked functional module; The source functional module is specifically used to determine the target on-chip sub-network according to the reachable paths of the functional modules that follow the data forwarding rules and are not blocked in each on-chip sub-network maintained in the routing table, wherein there is a target path of the functional modules that follow the data forwarding rules and are not blocked in the target on-chip sub-network; the target path is the path from the source functional module to the destination functional module.

2. The network on chip according to claim 1, characterized in that: The data forwarding rules configured for different on-chip sub-networks meet the following conditions: For any first data sent from a source functional module to a destination functional module, when there is a blocked functional module, there is at least one target on-chip sub-network that follows the configured data forwarding rules to complete the forwarding of the first data, and the forwarding path does not pass through the blocked functional module.

3. A chip, characterized in that: The invention comprises a functional module and a network on chip as claimed in any one of claims 1 to 2.

4. A data transmission method in a chip, characterized in that: The chip comprises: a functional module, an on-chip network, wherein the on-chip network comprises a functional module and at least two on-chip sub-networks; different data forwarding rules are pre-configured for different on-chip sub-networks; each of the functional modules is connected to a routing module belonging to the first on-chip sub-network and the second on-chip sub-network respectively; The on-chip sub-network is a mesh topology structure; The data forwarding rule is used to limit the direction of data forwarding by the routing module so that the data transmission of the on-chip sub-network is deadlock-free; the direction includes: from a first direction to a second direction, and from a second direction to a first direction; wherein the first direction is the horizontal direction of the mesh topology structure; and the second direction is the vertical direction of the mesh topology structure; The on-chip network includes a first on-chip sub-network and a second on-chip sub-network; For a first routing module and a second routing module connected to the same functional module, a data forwarding rule of the first on-chip sub-network restricts the redirection of the first routing module, which is different from a data forwarding rule of the second on-chip sub-network restricting the redirection of the second routing module; wherein the first routing module belongs to the first on-chip sub-network, and the second routing module belongs to the second on-chip sub-network; The method comprises: The source functional module determines the target on-chip sub-network according to the routing table; the target on-chip sub-network has a target path that follows the data forwarding rule; the target path is the path from the source functional module to the destination functional module; the routing table maintains the reachable paths that follow the data forwarding rule in each on-chip sub-network; The source functional module sends data to the routing module connected to the source functional module and belonging to the target on-chip sub-network; the data carries a forwarding instruction generated based on the target path, so that the routing module of the target on-chip sub-network sends the data to the destination functional module according to the forwarding instruction; The functional module is also used to: forward data from the routing module connected to itself and belonging to the first on-chip sub-network to the routing module connected to itself and belonging to the second on-chip sub-network; When there is a blocked functional module in the chip; The routing table maintains reachable paths in each on-chip sub-network that follow data forwarding rules and do not pass through blocked functional modules; The step of the source function module determining the target on-chip sub-network according to the routing table comprises: The source functional module determines the target on-chip sub-network according to the reachable paths of the functional modules that follow the data forwarding rules and are not blocked in each on-chip sub-network maintained in the routing table, wherein there is a target path in the target on-chip sub-network that follows the data forwarding rules and is not blocked by the functional modules; the target path is the path from the source functional module to the destination functional module.