An integrated circuit and its routing node
By introducing routing nodes and direct-connect switches into the integrated circuit, the switching between routing mode and shielding mode is realized, which solves the problem of inflexible on-chip network architecture and improves network flexibility and transmission efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG YUNHAI GUOCHUANG CLOUD COMPUTING EQUIP IND INNOVATION CENT CO LTD
- Filing Date
- 2026-03-13
- Publication Date
- 2026-06-30
Smart Images

Figure CN121833585B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of circuit technology, and in particular to an integrated circuit and its routing node. Background Technology
[0002] Network on-chip (NoC) is an on-chip communication network that integrates a large number of computing resources and connects these resources on an integrated circuit. See also: Figure 1 This is a schematic diagram of a common NoC network topology. R represents a routing node, and the required communication can be achieved by selecting a routing path.
[0003] As can be seen, the topology of a network-on-a-chip (NAT) defines the layout and connection methods of nodes and lines in the network. As one of the key technologies of NAT, it has a crucial impact on the overall network performance, including throughput, latency, fault tolerance, and load balancing. However, existing NAT architectures are not flexible enough.
[0004] In conclusion, how to effectively improve the flexibility of on-chip networks is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] The purpose of this invention is to provide an integrated circuit and its routing node to effectively improve the flexibility of on-chip networks.
[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:
[0007] In a first aspect, the present invention provides a routing node, which is disposed in an integrated circuit and, together with other routing nodes in the integrated circuit, constitutes the network topology of the integrated circuit. The routing node includes: a routing module, a network interface connected to the routing module, and a system-level functional module connected to the network interface; the routing module includes:
[0008] A routing unit connected to each input signal terminal and each output signal terminal of the routing node for signal routing;
[0009] N direct-connect switches; where N represents the total number of non-local input signal terminals of the routing node; each of the N direct-connect switches corresponds one-to-one with one of the N non-local input signal terminals of the routing node, and the first terminal of each of the N direct-connect switches is connected to its corresponding non-local input signal terminal; the second terminal of any one of the direct-connect switches is connected to the output signal terminal of the routing node according to a preset connection relationship;
[0010] The direct connection control unit is used to control all N direct connection switches to be turned off when the routing node is in routing mode, and to control all N direct connection switches to be turned on when the routing node is in shielding mode.
[0011] In one embodiment, the network topology is a mesh network topology, and the routing node is any one of the routing nodes in the network topology that has four neighboring routing nodes.
[0012] In one embodiment, the routing module includes four direct-connect switches;
[0013] The first terminal of the first direct-connect switch in the routing module is connected to the eastward input signal terminal of the routing node; the second terminal of the first direct-connect switch in the routing module is connected to the westward output signal terminal of the routing node.
[0014] The first terminal of the second direct-connect switch in the routing module is connected to the westward input signal terminal of the routing node; the second terminal of the second direct-connect switch in the routing module is connected to the eastward output signal terminal of the routing node.
[0015] The first terminal of the third direct-connect switch in the routing module is connected to the northward input signal terminal of the routing node; the second terminal of the third direct-connect switch in the routing module is connected to the southward output signal terminal of the routing node.
[0016] The first terminal of the fourth direct-connect switch in the routing module is connected to the south-facing input signal terminal of the routing node; the second terminal of the fourth direct-connect switch in the routing module is connected to the north-facing output signal terminal of the routing node.
[0017] In one embodiment, the network topology is a ring network topology, the routing node is any one routing node in the network topology, and the routing module in the routing node includes four direct-connect switches, so that any non-local input signal terminal of the routing node can be connected to the opposite non-local output signal terminal through the corresponding direct-connect switch.
[0018] In one embodiment, the routing unit includes:
[0019] Multiple input channels are used to connect one-to-one with each input signal terminal of the routing node;
[0020] Multiple output channels are used to connect one-to-one with each output signal terminal of the routing node;
[0021] A cross switch connected to each of the input channels and each of the output channels;
[0022] The routing calculation subunit is used for calculating routing paths;
[0023] A cross switch distributor for controlling the cross switch to perform route path control based on the route path calculation results of the route calculation subunit.
[0024] In one implementation, each input channel in the routing unit includes multiple virtual channels;
[0025] The routing unit also includes a virtual channel allocator for performing virtual channel allocation.
[0026] In one embodiment, the integrated circuit is an integrated circuit packaged from multiple chips, each chip including at least one routing node, and adjacent chips are connected through a serial interface or a parallel interface.
[0027] In one embodiment, the packaging method between adjacent chips is a standard package, any one of silicon interposer packaging, through-silicon via packaging, or silicon bridge packaging.
[0028] In one embodiment, the network topology is a tree-shaped network topology, and the routing node is any one routing node that is neither a top-level node nor a bottom-level node in the network topology.
[0029] The routing module in the routing node includes a direct-connect switch. The first end of the direct-connect switch is connected to the parent input signal terminal of the routing node, and the second end of the direct-connect switch is connected to the Nth child output signal terminal of the routing node.
[0030] The routing module in the routing node also includes M switching switches. The first end of any switching switch is connected to the parent input signal terminal of the routing node, and the second end of the i-th switching switch is connected to the i-th child output signal terminal of the routing node. i is a positive integer and 1≤i≤M, where M is the number of child output signal terminals of the routing node minus one.
[0031] The direct connection control unit is specifically used to: when the routing node is in shielded mode, control one of the direct connection switch and M switching switches to be turned on based on the shielding command; and control all of the direct connection switch and M switching switches to be turned off when the routing node is in routing mode.
[0032] In a second aspect, the present invention provides an integrated circuit including the routing node as described above.
[0033] By applying the technical solution provided in the embodiments of the present invention, the routing node can realize the direct connection function by adding a direct connection switch, which can effectively enhance the reconfiguration capability of the integrated circuit topology.
[0034] Specifically, the routing node is located within the integrated circuit and, together with other routing nodes within the integrated circuit, constitutes the network topology of the integrated circuit. The routing node includes: a routing module, a network interface connected to the routing module, and a system-level functional module connected to the network interface. The routing module includes routing units connected to each input and output signal terminal of the routing node for signal routing. In routing mode, the direct-connect control unit controls N direct-connect switches to be turned off, ensuring that the routing function of the routing unit is not affected. In other words, at this time, the routing node can effectively achieve the required routing. This application's solution sets N direct-connect switches in the routing module, where N represents the total number of non-local input signal terminals of the routing node. N direct-connect switches correspond one-to-one with the N non-local input signal terminals of the routing node, and the first terminal of each of the N direct-connect switches is connected to its corresponding non-local input signal terminal. The second terminal of any one of the direct-connect switches is connected to the output signal terminal of the routing node according to a preset connection relationship. Therefore, it can be seen that when the routing node is in shielded mode, with the direct-connect control unit controlling all N direct-connect switches to be turned on, the routing unit is essentially bypassed. At this time, the input signal is transmitted through the direct-connect switches of the routing node, thus avoiding passing through various circuits inside the routing unit, which can effectively save transmission time and reduce transmission latency. Of course, since the transmission is achieved using direct-connect switches, the signal transmission path is determined. Therefore, in practical applications, it is necessary to select whether to enable routing mode or shielded mode according to business requirements.
[0035] In summary, the routing node in this application has both routing and shielding modes. In shielding mode, the signal transmission path is fixed, but transmission time can be effectively saved and transmission latency reduced. By enabling the direct-connect switch, the network topology is changed, thus improving the flexibility of the integrated circuit. Attached Figure Description
[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0037] Figure 1 A schematic diagram of a traditional NoC network topology;
[0038] Figure 2 This is a schematic diagram of the structure of a routing node provided in a specific embodiment of the present invention;
[0039] Figure 3This is a schematic diagram of the structure of a routing unit provided in a specific embodiment of the present invention;
[0040] Figure 4 This is a schematic diagram of a mesh-based network topology in one specific embodiment of the present invention;
[0041] Figure 5 This is a schematic diagram of the connection relationship of the direct-connected switch of the routing module in a specific embodiment of the present invention;
[0042] Figure 6 This is a schematic diagram of the equivalent network topology after disabling 8 routing nodes in one specific embodiment of the present invention.
[0043] Figure 7 This is a schematic diagram of the equivalent network topology after one routing node is hidden in a specific embodiment of the present invention.
[0044] Figure 8 This is a schematic diagram of the equivalent network topology after one routing node is hidden in the tree network topology structure according to another specific embodiment of the present invention.
[0045] Figure 9 This is a schematic diagram illustrating the connection between adjacent cores via a serial or parallel interface in one specific embodiment of the present invention.
[0046] Figure 10 This is a schematic diagram illustrating the encapsulation of adjacent chips using a silicon interposer layer in one specific embodiment of the present invention.
[0047] Explanation of reference numerals in the attached diagram: 10-Routing module; 20-Network interface; 30-System-level functional module; 11-Routing unit; 12-Direct connection control unit; K1-First direct connection switch; K2-Second direct connection switch; K3-Third direct connection switch; K4-Fourth direct connection switch. Detailed Implementation
[0048] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.
[0049] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0050] The routing nodes in this application are located in the integrated circuit and together with other routing nodes in the integrated circuit, they constitute the network topology of the integrated circuit.
[0051] See also Figure 2 The routing node may include: a routing module 10, a network interface 20 connected to the routing module 10, and a system-level functional module 30 connected to the network interface 20. The routing module 10 includes:
[0052] A routing unit 11 is connected to each input signal terminal and each output signal terminal of a routing node and is used for signal routing.
[0053] There are N direct-connect switches; where N represents the total number of non-local input signal terminals of the routing node; each of the N direct-connect switches corresponds one-to-one with one of the N non-local input signal terminals of the routing node, and the first terminal of each of the N direct-connect switches is connected to its corresponding non-local input signal terminal; the second terminal of any one of the direct-connect switches is connected to the output signal terminal of the routing node according to a preset connection relationship.
[0054] The direct connection control unit 12 is used to control all N direct connection switches to be turned off when the routing node is in routing mode, and to control all N direct connection switches to be turned on when the routing node is in shielding mode.
[0055] Specifically, in this application, the routing node achieves direct connection functionality by adding a direct connection switch, which effectively enhances the topology reconfiguration capability of the integrated circuit.
[0056] Routing nodes are set in the integrated circuit and together with other routing nodes in the integrated circuit, they form the network topology of the integrated circuit. For example, the network topology includes mesh network topology, torus network topology, tree network topology (such as Fat Tree or other tree structures), butterfly network topology, etc.
[0057] A routing node may include a routing module 10, a network interface 20, and a system-level functional module 30, all connected by physical links. The network interface 20 connects the routing module 10 and the system-level functional module 30. The routing module 10 performs data routing. The system-level functional module 30 is the IP (Intellectual Property) core, the specific type of which can be configured according to actual needs. For example, it could be a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a dedicated IP core, a memory array, reconfigurable hardware, etc. Taking a CPU as an example, a single IP core can consist of one or more CPUs. Furthermore, it should be noted that the IP core types of different routing nodes within an integrated circuit can be the same or different.
[0058] The routing unit 11 is connected to each input signal terminal and each output signal terminal of the routing node. Figure 2 For ease of viewing, only the two input signal terminals (input signal terminal 1 and input signal 2) and two output signal terminals (output signal 1 and output signal 2) of the routing node are shown.
[0059] The routing unit 11 can perform signal routing, that is, it can receive input signals from any input signal terminal of the routing node where it is located, and output signals to any output signal terminal according to the routing path.
[0060] The specific structure of the routing unit 11 can be set and adjusted according to actual needs, as long as it can realize the routing function. For example, in a specific embodiment of the present invention, see [reference needed]. Figure 3 The routing unit 11 may include:
[0061] Multiple input channels are used to connect one-to-one with each input signal terminal of the routing node;
[0062] Multiple output channels are used to connect one-to-one with each output signal terminal of the routing node;
[0063] A cross switch that is connected to each input channel and each output channel;
[0064] The routing calculation subunit is used for calculating routing paths;
[0065] A cross switch distributor used to control cross switches for route path control based on the route path calculation results of the route calculation subunit.
[0066] Figure 3For ease of viewing and understanding, the diagram shows N input signal terminals in addition to the local input signal terminal. Therefore, the routing unit 11 shows N input channels, each connected to one of these N input signal terminals. Figure 3 These N input channels will be denoted as input channels 1 to N. Similarly, Figure 3 The diagram shows N output signal terminals besides the local output signal terminal. Therefore, the routing unit 11 shows N output channels, each connected to one of these N output signal terminals. Figure 3 These N output channels are denoted as output channels 1 to N.
[0067] The crossbar switch is connected to all N input channels and N output channels. It's understandable that if local input and output channels are shown, the crossbar switch will also be connected to them. However, it's important to note that for local input signals, a direct-connect switch is not required. This is because the direct-connect switch is enabled in shielded mode, effectively shielding the routing node. Therefore, it's unnecessary to receive local data from the routing node or output data to it.
[0068] The routing calculation subunit needs to calculate the routing path according to a preset routing algorithm. The specific routing algorithm used can be selected according to actual needs, and this application will not elaborate on this. The crossbar switch distributor can control the crossbar switch based on the routing path calculation result of the routing calculation subunit, thereby realizing the control of the routing path. For example, in one example, after controlling the crossbar switch according to the routing path calculation result, the input signal 1 from the first input signal terminal will be sent to the output channel 3, and thus output through the third output signal terminal of the routing node.
[0069] In one specific embodiment of the present invention, each input channel in the routing unit 11 includes multiple virtual channels, and correspondingly, the routing unit 11 further includes a virtual channel allocator for allocating virtual channels.
[0070] This implementation takes into account that if each input channel is a single channel, data from the same input signal end shares the same physical channel and input buffer, which can easily cause congestion and increase transmission latency. Therefore, in this implementation, several virtual channels can be set for each input channel. Each virtual channel can be associated with an independent buffer or other resource quota to achieve parallel transmission and reduce the probability of congestion. Of course, since virtual channels are set in this implementation, the routing unit 11 also needs to be equipped with a virtual channel allocator for allocating virtual channels. Naturally, the specific virtual channel allocation rules can be set and adjusted according to actual needs to effectively achieve the rational utilization of each virtual channel.
[0071] The system-level functional module 30 of this application, i.e., the IP core, can be categorized by function as a CPU core, GPU core, or dedicated function IP core, and by instruction set architecture as x86, ARM, etc. In one specific embodiment of this invention, the system-level functional module 30 can preferably be a system-level functional module 30 with extensible instructions, such as a RISC-V architecture IP core. RISC-V architecture IP cores have the advantages of being open source and having easily extensible instructions. Because instructions can be extended, dedicated topology communication calculation instructions can be designed for the IP core according to actual needs and the actual network topology. This allows for dedicated routing communication calculations based on the actual network topology, effectively accelerating communication calculations, reducing on-chip network latency, and improving communication efficiency.
[0072] The routing unit 11 has been described in detail above. It can be seen that when the routing node is in routing mode, the routing unit 11 can function normally because the direct connection control unit 12 controls all N direct connection switches to be turned off. In other words, at this time, the routing node is a routing node with complete routing functions. The shielded mode will be described in detail below.
[0073] As described above, the direct connection switch is enabled in shielded mode, which is equivalent to shielding the routing node. Therefore, in shielded mode, it is not necessary to receive local data from the routing node, nor is it necessary to output data to the local routing node. Therefore, the number of direct connection switches is the total number of non-local input signal terminals of the routing node. Since the local input signal terminal of the routing node is usually 1, the number of direct connection switches N is the total number of input signal terminals of the routing node minus 1.
[0074] N direct-connect switches correspond one-to-one with the N non-local input signal terminals of the routing node. The first terminal of each direct-connect switch is connected to its corresponding non-local input signal terminal, while the second terminal of each direct-connect switch can be connected to one of the output signal terminals of the routing node according to a preset connection relationship.
[0075] As can be seen, when the routing node is in shielded mode, all N direct-connect switches are turned on, which bypasses the routing unit 11. At this time, the input signal of any non-local input signal terminal of the routing node will be directly transmitted to a certain output signal terminal through the corresponding direct-connect switch. Therefore, it can avoid passing through various circuits inside the routing unit 11, which can effectively save transmission time.
[0076] Once all N direct-connect switches are turned on, the network topology changes, meaning that the routing path for that routing node is now fixed. Of course, the specific output signal terminal of the routing node connected to the second terminal of each direct-connect switch can be pre-defined as needed.
[0077] In one specific embodiment of the present invention, the network topology is a mesh network topology, and the routing node is any one routing node in the network topology that has 4 neighboring routing nodes.
[0078] The mesh network topology is considered in this implementation because it is a commonly used network topology in practical applications. Therefore, the solution proposed in this application can be applied to a mesh structure. Furthermore, this implementation considers that in a mesh structure, each internal routing node has a degree of 4, meaning each internal routing node connects to four neighbors (up, down, left, and right). Each edge routing node connects to 2-3 neighbors. Due to the lower degree of edge routing nodes, their routing paths are relatively simpler compared to internal routing nodes, eliminating the need for direct connection switches. In other words, for a mesh structure, this implementation applies the direct connection switch solution of this application to any one internal routing node, which has four neighboring routing nodes. Of course, in other implementations, if necessary, a corresponding number of direct connection switches can be configured for the edge routing nodes in the mesh structure.
[0079] See also Figure 4 This is a schematic diagram of a mesh network topology in one specific implementation. In this example, it is an 8×8 mesh structure, specifically an on-chip network composed of 64 RISC-V architecture CPUs. Figure 4In the diagram, each circle represents a routing node, including chiplets 1 to 4. Straight lines represent the connections between routing nodes within a chiplet, and dashed lines represent the encapsulation connections between chipslets (for example, the Chiplet encapsulation described later). Figure 4 The entire on-chip network consists of 4 cores, each of which is a 4×4 mesh structure on-chip network.
[0080] In one specific embodiment of the present invention, for a routing node with 4 neighboring routing nodes in a mesh structure, the routing module 10 of the routing node may include 4 direct-connect switches.
[0081] Specifically, the first terminal of the first direct-connect switch in the routing module 10 is connected to the eastward input signal terminal of the routing node; the second terminal of the first direct-connect switch in the routing module 10 is connected to the westward output signal terminal of the routing node.
[0082] The first terminal of the second direct-connect switch in the routing module 10 is connected to the westward input signal terminal of the routing node; the second terminal of the second direct-connect switch in the routing module 10 is connected to the eastward output signal terminal of the routing node.
[0083] The first terminal of the third direct-connect switch in the routing module 10 is connected to the northward input signal terminal of the routing node; the second terminal of the third direct-connect switch in the routing module 10 is connected to the southward output signal terminal of the routing node.
[0084] The first terminal of the fourth direct-connect switch in the routing module 10 is connected to the south-facing input signal terminal of the routing node; the second terminal of the fourth direct-connect switch in the routing module 10 is connected to the north-facing output signal terminal of the routing node.
[0085] For easier understanding, please refer to the following: Figure 5 This is a schematic diagram of the connection relationship of the direct-connection switch of the routing module 10 in one specific embodiment. Figure 5 In the routing module 10, the first direct-connect switch is denoted as K1, the second direct-connect switch as K2, the third direct-connect switch as K3, and the fourth direct-connect switch as K4.
[0086] In this implementation, it can be clearly seen from the connection relationship that if K1, K2, K3 and K4 are all on, then the routing unit 11 is bypassed, and the routing node is equivalent to having its routing function blocked. At this time, when the routing node receives the east input signal through the east input signal terminal, it will directly send it to the west output signal terminal through K1. That is, the routing node will send the west output signal to its west neighboring routing node.
[0087] Accordingly, when this routing node receives a west input signal through its west input signal terminal, it will directly transmit it to its east output signal terminal via K2, meaning it will send an east output signal to its neighboring routing node to its east. Similarly, when this routing node receives a north input signal through its north input signal terminal, it will directly transmit it to its south output signal terminal via K3, meaning it will send a south output signal to its neighboring routing node to its south. Likewise, when this routing node receives a south input signal through its south input signal terminal, it will directly transmit it to its north output signal terminal via K4, meaning it will send a north output signal to its neighboring routing node to its north.
[0088] This implementation sets up four direct connection switches for a routing node with four neighboring routing nodes in the mesh structure. When the four direct connection switches are turned on, it is equivalent to achieving direct connection in the north-south direction and direct connection in the east-west direction.
[0089] See also Figure 6 This is an equivalent network topology structure after disabling 8 routing nodes in one specific implementation. It can be seen that compared to... Figure 4 , Figure 6 In the example, each core has two routing nodes in shielded mode, at which point the routing node achieves direct connection in the north-south direction and direct connection in the east-west direction.
[0090] Furthermore, it should be noted that whether the routing node is in routing mode or shielded mode can be selected based on the actual business requirements. Figure 4 Taking the second row of chip 1 as an example, the second routing node from the left in the second row is a CPU routing node, the third routing node from the left in the second row is a GPU routing node, and the fourth routing node from the left in the second row is a storage routing node. For example, when performing tasks such as model inference, this GPU routing node needs to function normally, sending and receiving data through the network topology; therefore, this GPU routing node needs to be in routing mode. However, if a business change temporarily eliminates the need for model inference, causing this GPU routing node to stop sending and receiving data through the network topology, then the CPU routing node in this example might need to perform a large number of read and write accesses to the storage routing node. In this case, the GPU routing node can be set to a disabled mode. Figure 6 In the example, the GPU routing node is in shielded mode, which realizes direct connection in the north-south direction and direct connection in the east-west direction. This means that the GPU routing node is equivalent to a wire in the east-west direction, avoiding the delay caused by the internal circuit structure of the routing unit 11. Therefore, the CPU routing node in this example can realize read and write access to the storage routing node more quickly.
[0091] In addition to business changes, in practical applications, when a fault is detected in routing unit 11 of a routing node, it can also be set to shield mode, so that the faulty routing node does not affect data transmission, and a direct transmission path can be achieved through the direct connection switch in the faulty routing node.
[0092] In one specific embodiment of the present invention, the network topology is a ring network topology, the routing node is any one routing node in the network topology, and the routing module 10 in the routing node includes 4 direct-connect switches, so that any non-local input signal terminal of the routing node can be connected to the opposite non-local output signal terminal through the corresponding direct-connect switch.
[0093] The ring network topology, also known as the Torus structure, is an improved version of the Mesh structure. As mentioned above, internal routing nodes in a Mesh have four neighboring nodes, while edge routing nodes have only two to three neighboring nodes. The Torus structure connects the first and last nodes of a row and the first and last nodes of a column in a ring, making each node geometrically equivalent (all nodes have a degree of 4, meaning each routing node has four neighboring routing nodes). Therefore, in this implementation, for the Torus structure, each routing module 10 in each routing node can include four direct-connect switches. This allows any non-local input signal terminal of the routing node to be connected to the opposite non-local output signal terminal through the corresponding direct-connect switch, thereby realizing the north-south and east-west direct connections described above.
[0094] Tree-shaped network topologies are also commonly used, therefore the solution proposed in this application can also be applied to tree-shaped network topologies. For example... Figure 7 On the left is a three-layer tree-shaped network topology. Of course, in other specific implementations, it can also be other types of tree-shaped network topologies, such as Fat Tree structures.
[0095] It can be seen that, Figure 7 In the example, with all routing nodes in routing mode, complete routing functionality can be achieved across the entire tree-like network topology. See also... Figure 7 On the right, once the shielding switch of the left-side routing node in the second layer is turned on, the routing node can achieve direct connection. However, because in a tree-like network topology, a routing node has only one non-local input signal terminal (i.e., only one parent input signal terminal) and two or more non-local output signal terminals (i.e., two or more child output signal terminals), the specific output signal terminal to which the shielding switch is turned on to achieve direct connection can be preset according to the actual situation. Figure 7 In the example, when the shielding switch achieves a direct connection, it is specifically connected to the first sub-output signal terminal, thereby directly connecting to the first sub-node of the third layer. Of course, as described in the implementation below, a more flexible direct connection can be achieved by adding M switching switches.
[0096] Specifically, in one embodiment of the present invention, the network topology is a tree-shaped network topology, and the routing node is any one routing node that is neither the top-level node nor the bottom-level node in the network topology.
[0097] The routing module 10 in the routing node includes a direct-connect switch. The first end of the direct-connect switch is connected to the parent input signal terminal of the routing node, and the second end of the direct-connect switch is connected to the Nth child output signal terminal of the routing node.
[0098] The routing module 10 in the routing node also includes M switching switches. The first end of any switching switch is connected to the parent input signal terminal of the routing node, and the second end of the i-th switching switch is connected to the i-th child output signal terminal of the routing node. i is a positive integer and 1≤i≤M, and M is the number of child output signal terminals of the routing node minus one.
[0099] The direct connection control unit 12 is specifically used to: when the routing node is in shielded mode, control the direct connection switch and one of the M switching switches to be turned on based on the shielding command; and when the routing node is in routing mode, control the direct connection switch and the M switching switches to be turned off.
[0100] In this implementation, for a tree-shaped network topology, M switching switches are also provided to enhance the direct connection flexibility in shielded mode. The number M is one less than the number of child output signal terminals of the routing node. In other words, M switching switches and 1 direct connection switch can enable the connection between the parent input signal terminal of the routing node and any one child output signal terminal in shielded mode.
[0101] by Figure 8 For example, when the left routing node of the second layer is in shielded mode, the direct connection control unit 12 will, based on the shielding command, turn on the direct connection switch and one of the M switching switches. In this example, M equals 2. Figure 8 The three dashed lines shown represent the parent input signal terminal of the routing node in shielded mode. It can be connected to any one of the child output signal terminals. Which child output terminal is connected depends on which of the direct-connect switch and M switching switches is turned on. It can be seen that compared to... Figure 7 In this implementation, for tree-shaped network topologies, the direct connection flexibility is enhanced in the case of direct connections.
[0102] The routing node in this application is a routing node in an integrated circuit. In one specific embodiment of the present invention, the integrated circuit is an integrated circuit packaged from multiple chips, each chip including at least one routing node, and adjacent chips are connected through a serial interface or a parallel interface.
[0103] This implementation takes into account that if all routing nodes are manufactured in a single integrated circuit with a fixed structure, the entire integrated circuit would become unusable if any one of the routing nodes were to fail. To address this, this implementation considers that the integrated circuit can be packaged from multiple chips. That is, multiple chips with independent functions are integrated into a single package using advanced packaging technology, forming a high-performance, low-cost integrated circuit. Furthermore, if a chip fails, only that chip needs to be replaced, without replacing the entire integrated circuit. Moreover, when multiple chips are packaged into an integrated circuit, the required number and corresponding functions of chips can be packaged into the integrated circuit according to actual needs, offering high flexibility. The chip packaging described in this implementation is typically a chiplet package.
[0104] In addition, see the following: Figure 9 Adjacent die-to-die connections can be made via either a serial or parallel interface. Serial interfaces are preferred for applications requiring high bandwidth, long-distance communication, or where pin resources are expensive and the number of pins is limited. Parallel interfaces are preferred for applications requiring short-distance, high-efficiency communication and low latency.
[0105] In one specific embodiment of the present invention, the packaging method between adjacent chips can be either standard packaging or advanced packaging, such as any one of silicon interposer packaging, through-silicon via (TSV) packaging, or silicon bridge packaging. In practical applications, the interface and packaging method can be selected comprehensively based on different requirements such as chip application scenarios, performance, cost, power consumption, and supply chain.
[0106] For example, standard packaging can be used when bandwidth requirements are low, signal integrity requirements are not stringent, and low cost is a priority. Advanced packaging is used for scenarios requiring high bandwidth, low latency, and high performance. Advanced packaging methods include silicon interposer packaging, through-silicon via (TSV) packaging, and silicon bridge packaging. (See also...) Figure 10 This involves using silicon interposer technology for packaging.
[0107] Corresponding to the above embodiments of routing nodes, this embodiment of the invention also provides an integrated circuit that may include routing nodes as described in any of the above embodiments, and can be referred to in correspondence with the above.
[0108] By applying the technical solution provided in the embodiments of the present invention, the routing node can realize the direct connection function by adding a direct connection switch, which can effectively enhance the reconfiguration capability of the integrated circuit topology.
[0109] Specifically, the routing node is located within the integrated circuit and, together with other routing nodes within the integrated circuit, constitutes the network topology of the integrated circuit. The routing node includes: a routing module 10, a network interface 20 connected to the routing module 10, and a system-level functional module 30 connected to the network interface 20. The routing module 10 includes a routing unit 11 connected to each input signal terminal and each output signal terminal of the routing node for signal routing. In routing mode, the direct-connect control unit 12 controls N direct-connect switches to be turned off, ensuring that the routing function of the routing unit 11 is not affected. In other words, at this time, the routing node can effectively achieve the required routing. This application's solution sets N direct-connect switches in the routing module 10, where N represents the total number of non-local input signal terminals of the routing node. N direct-connect switches correspond one-to-one with the N non-local input signal terminals of the routing node, and the first terminal of each of the N direct-connect switches is connected to its corresponding non-local input signal terminal. The second terminal of any one of the direct-connect switches is connected to the output signal terminal of the routing node according to a preset connection relationship. Therefore, it can be seen that when the routing node is in shielded mode, with the direct-connect control unit 12 controlling all N direct-connect switches to be turned on, the routing unit 11 is essentially bypassed. At this time, the input signal is transmitted through the direct-connect switches of the routing node, thus avoiding passing through various circuits inside the routing unit 11, which can effectively save transmission time and reduce transmission delay. Of course, since the transmission is achieved using direct-connect switches, the signal transmission path is determined. Therefore, in practical applications, it is necessary to select whether to enable routing mode or shielded mode according to business requirements.
[0110] In summary, the routing node in this application has both routing and shielding modes. In shielding mode, the signal transmission path is fixed, but transmission time can be effectively saved and transmission latency reduced. By enabling the direct-connect switch, the network topology is changed, thus improving the flexibility of the integrated circuit.
[0111] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0112] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.
[0113] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.
[0114] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only intended to help understand the technical solutions and core ideas of the present invention. It should be noted that those skilled in the art can make several improvements and modifications to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the protection scope of the present invention.
Claims
1. A routing node, characterized by The routing node is located within the integrated circuit and, together with other routing nodes within the integrated circuit, constitutes the network topology of the integrated circuit. The routing node includes: a routing module, a network interface connected to the routing module, and a system-level functional module connected to the network interface; the routing module includes: A routing unit connected to each input signal terminal and each output signal terminal of the routing node for signal routing; N direct-connect switches; where N represents the total number of non-local input signal terminals of the routing node; each of the N direct-connect switches corresponds one-to-one with one of the N non-local input signal terminals of the routing node, and the first terminal of each of the N direct-connect switches is connected to its corresponding non-local input signal terminal; the second terminal of any one of the direct-connect switches is connected to the output signal terminal of the routing node according to a preset connection relationship; The direct connection control unit is used to control all N direct connection switches to be turned off when the routing node is in routing mode, and to control all N direct connection switches to be turned on when the routing node is in shielding mode. The network topology is a tree-shaped network topology, and the routing node is any one routing node that is neither a top-level node nor a bottom-level node in the network topology. The routing module in the routing node includes a direct-connect switch. The first end of the direct-connect switch is connected to the parent input signal terminal of the routing node, and the second end of the direct-connect switch is connected to the Nth child output signal terminal of the routing node. The routing module in the routing node also includes M switching switches. The first end of any switching switch is connected to the parent input signal terminal of the routing node, and the second end of the i-th switching switch is connected to the i-th child output signal terminal of the routing node. i is a positive integer and 1≤i≤M, where M is the number of child output signal terminals of the routing node minus one. The direct connection control unit is specifically used to: when the routing node is in shielded mode, control one of the direct connection switch and M switching switches to be turned on based on the shielding command; and control all of the direct connection switch and M switching switches to be turned off when the routing node is in routing mode.
2. The routing node of claim 1, wherein, The routing unit includes: Multiple input channels are used to connect one-to-one with each input signal terminal of the routing node; Multiple output channels are used to connect one-to-one with each output signal terminal of the routing node; A cross switch connected to each of the input channels and each of the output channels; The routing calculation subunit is used for calculating routing paths; A cross switch distributor for controlling the cross switch to perform route path control based on the route path calculation results of the route calculation subunit.
3. The routing node of claim 2, wherein, Each input channel in the routing unit includes multiple virtual channels; The routing unit also includes a virtual channel allocator for performing virtual channel allocation.
4. The routing node of claim 1, wherein, The integrated circuit is an integrated circuit packaged from multiple chips, each chip including at least one routing node, and adjacent chips are connected through a serial interface or a parallel interface.
5. The routing node of claim 4, wherein, The packaging method between adjacent chips is standard packaging, which can be any one of silicon interposer packaging, through-silicon via packaging, or silicon bridge packaging.
6. An integrated circuit, characterized by Includes the routing node as described in any one of claims 1 to 5.