Network topology and route optimization design method oriented to Chiplet core particle interconnection
The network topology and routing optimization design for chiplet interconnection addresses high interconnection costs by determining input and output codes based on address information, optimizing switch connections to reduce chiplet interface area and costs.
Patent Information
- Application Number
- CN202411227936.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-03
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-09-03
AI Technical Summary
The increasing cost of designing and manufacturing monolithic large-scale multi-core chips due to high interconnection costs between split cores, exacerbated by core interface area overhead, necessitates a reduction in interconnection costs between different chiplets.
A network topology and routing optimization design method for chiplet interconnection that determines input and output codes based on address information, utilizing a preset output mutually exclusive pair relationship to control switch connections, reducing the complexity of calculations and hardware resources, thereby minimizing the chiplet interface area and interconnection costs.
This method significantly reduces the interconnection cost between chiplets by optimizing the routing algorithm and network topology, minimizing hardware resources and interface area, making it suitable for hardware implementation.
Smart Images

Figure CN119094424A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of chip technology, and in particular to a network topology and routing optimization design method for chiplet interconnection. Background Art
[0002] As the cost of transistors slows down and manufacturing yield challenges grow, the cost of designing and manufacturing monolithic large-scale multi-core chips has skyrocketed.
[0003] In the related art, in order to reduce the cost of a large-scale single-chip multi-core chip without affecting its function, the large-scale single-chip multi-core chip can be split into small cores, and the cores can be reused through packaging integration technology. However, the interconnection between the split cores needs to be realized through the core interface, which increases the core interface area overhead compared to a large single chip, resulting in an increase in the interconnection cost between different cores.
[0004] Therefore, how to reduce the interconnection cost between different core particles becomes a technical problem that needs to be solved urgently. Summary of the invention
[0005] Based on this, it is necessary to provide a network topology and routing optimization design method for chiplet interconnection that can reduce the interconnection cost between different chiplets in response to the above technical problems.
[0006] In a first aspect, the present application provides a network topology and routing optimization design method for chiplet interconnection, which is applied to chiplets and includes:
[0007] Determine the input code and output code of the data to be exchanged in the core grain interface of the core grain according to the address information of the data to be exchanged;
[0008] Determine control codes of multiple switches of a chiplet interconnection network in a chiplet interface based on a preset output mutually exclusive pair relationship, an input code, and an output code;
[0009] The control code of each switch is used to control the interconnection of the data to be exchanged between adjacent core particles.
[0010] In one embodiment, the chiplet interconnect network includes an input switch group, an output switch group, and a plurality of sub-network topologies, wherein the plurality of sub-network topologies are connected to the input switch group and the output switch group through inter-level connections;
[0011] Based on the preset output mutually exclusive pair relationship, input coding and output coding, control codes of multiple switches of the chiplet interconnection network in the chiplet interface are determined, including:
[0012] Based on the preset output mutually exclusive pair relationship, the input code and the output code, obtaining the first control code of the plurality of switches in the input switch group and the output switch group;
[0013] According to the first control code and the inter-level connection, second control codes of multiple switches in multiple sub-network topologies are determined.
[0014] In one embodiment, based on a preset output mutually exclusive pair relationship, an input code, and an output code, obtaining first control codes of a plurality of switches in an input switch group and an output switch group includes:
[0015] Based on the output mutually exclusive pair relationship, the input code and the output code, the target connection mode inside all switches in the output switch group and the input switch group is obtained;
[0016] Based on the target connection mode, first control codes of multiple switches in the output switch group and the input switch group are determined.
[0017] In one embodiment, the two input codes of each input switch in the input switch group are an input mutually exclusive pair;
[0018] Based on the output exclusive pair relationship, input code, and output code, the target connection mode inside all switches in the input switch group and the output switch group is obtained, including:
[0019] Based on the output mutually exclusive pair relationship, the number of output mutually exclusive pairs of any set of input codes in the reference set of mutually exclusive points is obtained;
[0020] Based on the number of output mutually exclusive pairs, the target connection mode within all switches in the input switch group and the output switch group is determined.
[0021] In one embodiment, based on the number of output mutually exclusive pairs, determining the target connection mode inside all switches in the input switch group and the output switch group includes:
[0022] Based on the number of output mutually exclusive pairs of any set of input codes in the reference mutually exclusive point set, an initial connection mode inside the switch corresponding to the input codes is adjusted to obtain a candidate connection mode;
[0023] Based on the number of mutually exclusive pairs of each output, each candidate connection mode is adjusted twice to obtain the target connection mode inside all switches.
[0024] In one of the embodiments, the chiplet interface of the chiplet includes a chiplet interconnection network, the network topology of the chiplet interconnection network is composed of an input switch group, an output switch group and multiple sub-network topologies, any sub-network topology includes an input sub-switch group and an output sub-switch group, the input sub-switch group includes multiple groups of adjacent input switches, and an intra-layer channel is set between each group of adjacent input switches; the intra-layer channel is used for data interaction between adjacent input switches.
[0025] In one embodiment, each of the adjacent input switches includes an inter-layer interface and an intra-layer interface, wherein the inter-layer interface is connected to a plurality of switches of the output sub-switch group; and the intra-layer interface is used to connect to another switch of the adjacent input switches.
[0026] In a second aspect, the present application also provides a network topology and routing optimization design device for chiplet interconnection, including:
[0027] A coding determination module, used for determining the input coding and output coding of the data to be exchanged in the core-grain interface of the core-grain according to the address information of the data to be exchanged;
[0028] A control code determination module, used to determine the control codes of multiple switches of the chiplet interconnection network in the chiplet interface based on a preset output mutually exclusive pair relationship, an input code, and an output code;
[0029] The control module is used to control the interconnection of the data to be exchanged between adjacent core particles through the control code of each switch.
[0030] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the content of any one of the embodiments of the network topology and routing optimization design method for chiplet interconnection in the first aspect is implemented.
[0031] In a fourth aspect, the present application further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the content of any one of the embodiments of the network topology and routing optimization design method for chiplet interconnection in the first aspect above is implemented.
[0032] In a fifth aspect, the present application further provides a computer program product, including a computer program, which, when executed by a processor, implements the content of any one of the embodiments of the network topology and routing optimization design method for chiplet interconnection in the first aspect above.
[0033] The network topology and routing optimization design method for chiplet interconnection determines the input code and output code of the data to be exchanged in the chiplet interface of the chiplet according to the address information of the data to be exchanged; determines the control codes of multiple switches of the chiplet interconnection network in the chiplet interface based on the preset output mutually exclusive pair relationship, input code and output code; and controls the interconnection between adjacent chiplets through the control code of each switch. In the process of determining the control code, this method does not consider the specific content of the input data and output data in the chiplet interface, but analyzes the input number corresponding to the input data and the output number corresponding to the output data based on the preset output mutually exclusive pair relationship, which can greatly reduce the amount of calculation in the analysis process. Then, the area of the chiplet interface will also be reduced accordingly, thereby reducing the interconnection cost between different chiplets. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.
[0035] FIG1 is a diagram of an application environment of a network topology and routing optimization design method for chiplet interconnection in one embodiment;
[0036] FIG2 is a flow chart of a method for designing a network topology and routing optimization for chiplet interconnection in one embodiment;
[0037] FIG3 is a schematic diagram of input coding of an input switch in one embodiment;
[0038] FIG4 is a schematic diagram of outputting mutually exclusive pair relationships in one embodiment;
[0039] FIG5 is a schematic diagram of a flow chart of a network topology and routing optimization design method for chiplet interconnection in one embodiment;
[0040] FIG6 is a flow chart of a method for designing a network topology and routing optimization for chiplet interconnection in one embodiment;
[0041] FIG7 is a flow chart of a method for designing a network topology and routing optimization for chiplet interconnection in one embodiment;
[0042] FIG8 is a flow chart of a method for designing a network topology and routing optimization for chiplet interconnection in one embodiment;
[0043] FIG9 is a schematic diagram of a flow chart of a network topology and routing optimization design method for chiplet interconnection in one embodiment;
[0044] FIG10 is a schematic diagram of an 8×8 BENES network architecture in one embodiment;
[0045] FIG11 is a schematic diagram of data routing in a sub-network topology with congestion in one embodiment;
[0046] FIG12 is a schematic diagram of data routing in a sub-network topology with congestion in one embodiment;
[0047] FIG13 is a schematic diagram of data routing in a non-blocking sub-network topology according to an embodiment;
[0048] FIG14 is a schematic diagram of data routing in a non-blocking sub-network topology according to an embodiment;
[0049] FIG15 is a circuit diagram of a conventional 4×4 BENES network in one embodiment;
[0050] FIG16 is a circuit diagram of a 4×4 Butterfly network in one embodiment;
[0051] FIG17 is a schematic diagram of the output of a 4×4 Butterfly network in one embodiment;
[0052] FIG18 is a schematic diagram of an improved 8×8 BENES network architecture in one embodiment;
[0053] FIG. 19 is a structural block diagram of a network topology and routing optimization design device for chiplet interconnection in one embodiment. DETAILED DESCRIPTION
[0054] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0055] Before introducing the technical solution of the present application in detail, a brief description of the background technology of the present application is first given.
[0056] In order to provide high computing throughput and on-chip memory, the area of deep neural network (DNN) multi-core processors for artificial intelligence (AI) is getting larger and larger. However, facing the end of Moore's Law, the cost reduction rate of transistors has slowed down, and the challenge of manufacturing yield has become increasingly greater. As a result, the design and manufacturing costs of monolithic large-scale multi-core chips have skyrocketed.
[0057] Compared with a single large chip, the chiplet architecture technology can split a single large-scale multi-core chip into small chiplets and reuse the chiplets through advanced packaging integration technology. It has the advantages of shorter design and development cycle, lower manufacturing cost, and easier customization.
[0058] However, the interconnection between the split cores needs to be realized through the core interface. Compared with a single large chip, the core interface area overhead is increased, resulting in an increase in the interconnection cost between different cores. The on-chip interconnection network of the core interface cannot reduce the interconnection cost between different cores. For example, the on-chip interconnection network can be a wireless mesh network (MESH network) and a non-blocking network in a multi-level interconnection network (BENES network).
[0059] Taking the BENES network as an example, in the related art, the algorithm of the BENES network is mainly optimized by iterative method to reduce the chip interface area. Specifically, the algorithm first sets that in the same 2x2 switch, the two inputs are mutually exclusive, the two outputs are mutually exclusive, and the inputs or outputs of any different switches are not mutually exclusive. Starting from the leftmost and rightmost switches, (1) input the bit sequence to be replaced 1,2,3,...,N-1,N; output .definition is a permutation transformation relationship. According to the setting, we get (2) input mutually exclusive pair: ; (3) Output mutually exclusive pairs: ; (4) Construct two mutually exclusive point sets and each set contains elements that satisfy the condition , The specific operation is: in all P i Randomly select a P i From P i Select an element a as an element in X, ensuring that a belongs to O j , and then find O that satisfies the conditions j Next, select another element b other than a, and let b belong to P k Note that b cannot be selected as an element of X, but only from P k Select another element from as the element of X. For each P i ( ) is screened, and the cycle is repeated to obtain the set X. The set Y consists of elements that are not in X. (5) X and Y are used to determine the state of the two-stage 2x2 switches that are symmetrical in position, that is, to determine whether they are crossover or straight-through. For the first-stage 2×2 switch on the left, the elements in X connect the upper output end of the two output ends of the switch, and the elements in Y connect the lower output end of the two output ends of the switch. For the first-stage switch on the right, the elements in X connect the upper input end of the two input ends of the switch, and the elements in Y connect the lower input end of the two input ends of the switch, thereby determining the state of the left and right columns of switches. (6) Using the two columns of switches determined in step (5), the bits to be replaced pass through the switches and are sent to the input and output of the middle two-stage switches through the inter-stage connection. A new mutually exclusive pair is thus generated. Then steps (4) and (5) are repeated until the states of all switches are determined. In addition, the BENES network is an efficient switching network topology structure, which is usually used to build large-scale communication systems or computer networks. It is characterized by a compact structure and a simple routing algorithm, and has the characteristics of being rearrangeable and non-blocking. A BENES network is usually composed of a multi-stage switching network, where each stage has a set of 2x2 switches, each of which is responsible for exchanging data between input and output ports.
[0060] However, from the perspective of hardware implementation, the algorithm requires a large number of comparators to compare addresses when looking for mutually exclusive pairs of inputs and outputs. Moreover, the implementation process is complicated, and a large number of data selectors are required to select and process the results after each comparison, resulting in large area and power consumption overhead. At the same time, the simplest 4x4 BENES network can be regarded as an intermediate switch of a larger-scale BENES network. However, the traditional 4x4 BENES network has resource redundancy, which increases the area and power consumption overhead. In other words, the routing algorithm of the existing BENES network is not suitable for hardware programming design and implementation. Its logic has large area and power consumption overhead for designing dedicated hardware modules, which is not conducive to optimizing the core interface overhead required for interconnection between chiplets. Secondly, the traditional 4x4 BENES network currently used has resource redundancy in hardware implementation, and there is a certain amount of room for optimization.
[0061] Therefore, in response to the above problems, the present application provides a network topology and routing optimization design method for chiplet interconnection, which optimizes from two perspectives: the routing algorithm of the BENES network and the network topology of the chiplet interconnection network, and can reduce the interconnection cost between different chiplets. Of course, the technical solution provided in the embodiments of the present application is not limited to solving only the above problems, but also has other technical effects, which can be specifically described in the following embodiments. The technical solution of the present application is described in detail below.
[0062] The network topology and routing optimization design method for chiplet interconnection provided in the embodiment of the present application can be applied to the application environment shown in FIG1. The chip in the application environment can be a computer device, and the computer device can be a server, a personal computer, a laptop, a smart phone, a tablet computer, a smart mobile phone, etc. The computer device may include a processor, a memory, and a network interface connected via a system bus or wirelessly. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device may include a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data in the process of chiplet interconnection. The network interface of the computer device is used to communicate with an external terminal through a network connection, and the computer program is executed by the processor to implement a network topology and routing optimization design method for chiplet interconnection.
[0063] In an exemplary embodiment, as shown in FIG2 , a network topology and routing optimization design method for chiplet interconnection is provided, which is described by taking the method applied to the chiplet in FIG1 as an example, and includes the following steps 201 to 203. Among them:
[0064] S201, determining the input code and output code of the data to be exchanged in the chiplet interface of the chiplet according to the address information of the data to be exchanged.
[0065] The data to be exchanged refers to data that needs to be interconnected between adjacent core particles. Each core particle has a core particle interface, and the core particle interface is connected to the interface of its adjacent core particle by wired or wireless means to interconnect with its adjacent core particles.
[0066] When the data to be exchanged needs to be interconnected between the core particles, the address information carried by the data to be exchanged includes the core particles involved in the exchange process of the data to be exchanged, and the location of the core particle interconnection network in the core particle interface of each core particle. The address information of the data to be exchanged can be manually configured by the user according to the data routing requirements. Alternatively, the address information of the data to be exchanged can also be obtained by automatic allocation.
[0067] In an embodiment of the present application, after the core particle obtains the address information of the data to be exchanged, the core particle can analyze the address information to determine the output encoding method of the data to be exchanged. And based on the output encoding method, the output encoding of the data to be exchanged in the core particle interface of the core particle is determined. It should be noted that the input encoding in the core particle interface is a pre-determined encoding. For example, for an 8×8 core particle interconnection network, the core particle interconnection network includes 4 input switches, each of which includes two input encodings. Then, the encoding of the input switch of the core particle interconnection network can be 0, 1, 2, 3, 4, 5, 6 and 7.
[0068] It should be noted that, for any input switch on the core-grain interconnection network in the core-grain interface, the two input codes on the input switch are an input mutually exclusive pair. FIG3 is a schematic diagram of the input coding of the input switch, where X and Y are two mutually exclusive point sets. The figure includes four input switches, each of which includes an X value and a Y value. Each dotted box in the figure represents the input coding of an input switch. The four switches are G0, G1, G2, and G3. The input coding of the first switch G0 is 0 and 1, the input coding of the second switch G1 is 2 and 3, the input coding of the third switch G2 is 4 and 5, and the input coding of the fourth switch G4 is 6 and 7.
[0069] S202, determining control codes of a plurality of switches of a chiplet interconnection network in a chiplet interface based on a preset output mutually exclusive pair relationship, an input code, and an output code.
[0070] The preset output mutually exclusive pair relationship can be set in advance based on user needs. Figure 4 shows a schematic diagram of the output mutually exclusive pair relationship. It can be seen from the figure that 0 and 4 are an output mutually exclusive pair, 1 and 2 are an output mutually exclusive pair, 3 and 6 are an output mutually exclusive pair, and 5 and 7 are an output mutually exclusive pair.
[0071] The control code is an identification code indicating whether the internal connection relationship of the switch needs to be adjusted. For example, if the internal connection relationship of the switch does not need to be adjusted, the control code of the switch is 0; if the internal connection relationship of the switch needs to be adjusted, the control code of the switch is 1. The internal connection relationship of the switch includes straight-through connection and cross-connection.
[0072] In an embodiment of the present application, the core grain can obtain the number of output mutually exclusive pairs of any group of input codes in the reference set of points to be mutually exclusive based on the output mutually exclusive pair relationship. And use the number of each output mutually exclusive pair to determine whether the internal connection relationship of multiple switches in the core grain interconnection network in the core grain interface needs to be adjusted. Based on the judgment result, the control codes of multiple switches are determined. Alternatively, the core grain can also input the preset output mutually exclusive pair relationship, input code and output code into a preset control code determination model. The control code determination model uses the output mutually exclusive pair relationship to analyze the input code and the output code, and outputs the control codes of multiple switches in the core grain interconnection network in the core grain interface. The embodiment of the present application does not limit the specific method of determining the control codes of multiple switches in the core grain interconnection network in the core grain interface based on the preset output mutually exclusive pair relationship, input code and output code.
[0073] S203, controlling the interconnection of the data to be exchanged between adjacent core particles through the control code of each switch.
[0074] In the embodiment of the present application, after the control codes of multiple switches of the chiplet interconnection network in the chiplet interface are determined, the chiplet can determine the data exchange path of the data to be exchanged in the chiplet interconnection network based on the multiple control codes. Through the data exchange path, the data to be exchanged is transmitted between adjacent chiplets, thereby realizing the chiplet interconnection.
[0075] In the above-mentioned network topology and routing optimization design method for chiplet interconnection, the input code and output code of the data to be exchanged in the chiplet interface of the chiplet are determined according to the address information of the data to be exchanged; the control codes of multiple switches of the chiplet interconnection network in the chiplet interface are determined based on the preset output exclusive pair relationship, input code and output code; the data to be exchanged is controlled to be interconnected between adjacent chiplets through the control code of each switch. In the process of determining the control code, this method does not consider the specific content of the input data and output data in the chiplet interface, and analyzes the input number corresponding to the input data and the output number corresponding to the output data based on the preset output exclusive pair relationship, which can greatly reduce the amount of calculation in the analysis process, reduce the complexity of hardware resources, and reduce the area of the chiplet interface accordingly, thereby reducing the interconnection cost between different chiplets.
[0076] The chiplet interconnection network on the chiplet interface includes an input switch group, an output switch group and multiple sub-network topologies, and the multiple sub-network topologies are connected to the input switch group and the output switch group through inter-layer connections. In one embodiment, as shown in FIG5 , the specific content of the control code of multiple switches of the chiplet interconnection network in the chiplet interface determined based on the preset output mutually exclusive pair relationship, input coding and output coding includes:
[0077] S301, based on a preset output mutually exclusive pair relationship, an input code, and an output code, obtain first control codes of multiple switches in an input switch group and an output switch group.
[0078] The input switch group refers to the switch group at the outermost input layer in the chip interconnect network, and the output switch group refers to the switch group at the outermost output layer in the chip interconnect network, that is, both the input switch group and the output switch group are at the outermost layer of the chip interconnect network. For an 8×8 chip interconnect network, the chip interconnect network includes an input switch group, an output switch group, and two 4×4 sub-network topologies, and both the input switch group and the output switch group include 4 switches.
[0079] In the embodiment of the present application, the core particle can analyze the input code and the output code by using the preset output mutually exclusive pair relationship, and judge whether the internal connection relationship of the multiple switches in the input switch group and the output switch group needs to be adjusted. And according to the judgment result, the first control code of the multiple switches is determined.
[0080] S302: Determine second control codes of multiple switches in multiple sub-network topologies according to the first control code and the inter-layer connection.
[0081] In an embodiment of the present application, after obtaining the first control code of multiple switches in the outermost input switch group and output switch group, the core particle can determine the outermost input coding and output coding of the sub-network topology based on the inter-layer connection. Then, based on the outermost input coding and output coding of the sub-network topology and the output mutually exclusive pair relationship, determine the control code of multiple switches in the outermost input switch group and output switch group of the sub-network topology. If the sub-network topology is a 4×4 network topology, the control code of multiple switches in the input switch group and output switch group is used as the second control code of multiple switches in the sub-network topology.
[0082] In the above-mentioned network topology and routing optimization design method for chiplet interconnection, based on the preset output mutually exclusive pair relationship, input coding and output coding, the first control code of multiple switches in the input switch group and the output switch group is obtained; according to the first control code and the inter-level connection, the second control code of multiple switches in multiple sub-network topologies is determined. This method analyzes the input coding and the output coding through the output mutually exclusive pair relationship, and starts from the outer layer of the chiplet interconnection network, and obtains the control codes of multiple switches in the chiplet interconnection network step by step.
[0083] Next, the specific process of obtaining the first control code is introduced. As shown in FIG6 , the specific content of obtaining the first control code of multiple switches in the input switch group and the output switch group based on the preset output mutually exclusive pair relationship, input code and output code includes the following steps:
[0084] S401, based on the output mutually exclusive pair relationship, the input code and the output code, obtain the target connection mode inside all switches in the output switch group and the input switch group.
[0085] For any switch, the connection modes inside the switch include straight-through connection and cross-connection. The target connection mode is the interaction path of the data to be exchanged in the chiplet interconnection network during data interaction.
[0086] In the embodiment of the present application, any group of input codes is an input mutually exclusive pair. Then, for any group of input codes, the core particle can obtain the number of output mutually exclusive pairs of the group of input codes in the reference mutually exclusive point set based on the preset output mutually exclusive pair relationship. And based on the number of output mutually exclusive pairs, determine the first control code of the input switch and the output switch corresponding to the group of input codes. For other input codes, the first control code of the input switch and the output switch corresponding to other groups of input codes can also be obtained through the above method.
[0087] S402, determining first control codes of multiple switches in the output switch group and the input switch group based on the target connection mode.
[0088] In the embodiment of the present application, there is a mapping relationship between the connection mode inside the switch and the control code. After obtaining the connection modes of multiple input switches and output switches, for any switch, the core particle can analyze the target connection mode based on the mapping relationship and determine the control code corresponding to the target connection mode. The control code corresponding to the target connection mode of all switches is used as the first control code.
[0089] In the above-mentioned network topology and routing optimization design method for chiplet interconnection, based on the output mutually exclusive pair relationship, input coding and output coding, the target connection mode inside all switches in the output switch group and the input switch group is obtained; based on the target connection mode, the first control code of multiple switches in the output switch group and the input switch group is determined. This method can accurately analyze the input coding and the output coding through the output mutually exclusive pair relationship, determine the target connection mode inside all switches in the output switch group and the input switch group, and then accurately obtain the first control codes of multiple switches based on the target connection mode of all switches.
[0090] Assuming that the two input codes of each input switch in the input switch group are an input mutually exclusive pair, then, in one embodiment, as shown in FIG. 7 , the specific content of obtaining the target connection mode inside all switches in the input switch group and the output switch group based on the output mutually exclusive pair relationship, the input code and the output code includes the following steps:
[0091] S501, based on the output mutually exclusive pair relationship, obtaining the number of output mutually exclusive pairs of any group of input codes in the reference to-be-mutually-exclusive point set.
[0092] Continuing to refer to FIG. 4 , the reference to-be-mutually-exclusive point set refers to the mutually-exclusive point set X in FIG. 4 , that is, the reference to-be-mutually-exclusive point set includes 0, 2, 4, and 6.
[0093] In an embodiment of the present application, for any set of input codes, the core particle can obtain the output mutually exclusive codes corresponding to the two input codes, for example, when the input code is 0, the corresponding output mutually exclusive code is 4; when the input code is 1, the corresponding output mutually exclusive code is 2. And respectively determine whether the two output mutually exclusive codes are in the reference set of mutually exclusive points, and determine the number of output mutually exclusive pairs of the group of input codes in the reference set of mutually exclusive points according to the judgment result. For example, by respectively determining whether 0 and 4, 1 and 2 are in the reference set of mutually exclusive points, it can be determined that 0 and 4 are in the reference set of mutually exclusive points, and 1 and 2 are not in the reference set of mutually exclusive points. Therefore, the number of output mutually exclusive pairs of the group of input codes in the reference set of mutually exclusive points is 1.
[0094] S502: Determine the target connection mode inside all switches in the input switch group and the output switch group based on the number of output mutually exclusive pairs.
[0095] In an embodiment of the present application, after obtaining the number of output mutually exclusive pairs corresponding to any switch, the core particle can compare the number of output mutually exclusive pairs with a preset threshold. If the number of output mutually exclusive pairs is greater than or equal to the preset threshold, it is determined that the connection mode inside the switch needs to be adjusted, and the adjusted connection mode is used as the target connection mode inside the switch; if the number of output mutually exclusive pairs is less than the preset threshold, it is determined that the connection mode inside the switch does not need to be adjusted, and the default connection mode inside the switch is directly used as the target connection mode.
[0096] In the above-mentioned network topology and routing optimization design method for chiplet interconnection, based on the output mutually exclusive pair relationship, the number of output mutually exclusive pairs of any group of input codes in the reference set of mutually exclusive points is obtained; based on the number of output mutually exclusive pairs, the target connection mode inside all switches in the input switch group and the output switch group is determined. Based on the output mutually exclusive pair relationship, this method can accurately obtain the number of output mutually exclusive pairs of each input code in the reference set of mutually exclusive points, so that based on the number of output mutually exclusive pairs, the target connection mode inside all switches in the input switch group and the output switch group can be accurately determined.
[0097] Next, the specific content of determining the target connection mode inside all switches in the input switch group and the output switch group based on the number of output mutually exclusive pairs is introduced through an embodiment, as shown in FIG8 , and the specific content includes:
[0098] S601, based on the number of output mutually exclusive pairs of any group of input codes in the reference mutually exclusive point set, adjust the initial connection mode inside the switch corresponding to the input codes to obtain candidate connection modes.
[0099] In an embodiment of the present application, after obtaining the number of output mutually exclusive pairs of any set of input codes in the reference mutually exclusive point set, the core particle can compare the number of output mutually exclusive pairs with a preset threshold, and determine whether the initial connection mode inside the switch needs to be adjusted based on the comparison result. If it is determined that adjustment is required, the initial connection mode inside the switch is adjusted to obtain a candidate connection mode. For example, when the initial connection mode is a straight-through connection, the candidate connection mode is a cross connection; when the initial connection mode is a cross connection, the candidate connection mode is a straight-through connection.
[0100] S602: Based on the number of mutually exclusive pairs of outputs, each candidate connection mode is adjusted twice to obtain a target connection mode inside all switches.
[0101] In the embodiment of the present application, one adjustment is made based on the number of mutually exclusive pairs of outputs for each group of input codes, that is, the adjustment is made from a local perspective. After one adjustment is completed for all switches, in order to further ensure the accuracy of the connection method, the connection method can be further adjusted from a global perspective. Specifically, the core particle can comprehensively consider the number of mutually exclusive pairs of outputs and make a secondary adjustment to the candidate connection methods of each switch from an overall perspective. Through the secondary adjustment, under the target connection method inside all switches, the data to be exchanged will not be blocked when interacting in the core particle interconnection network.
[0102] In the above-mentioned network topology and routing optimization design method for chiplet interconnection, based on the number of output mutually exclusive pairs of any set of input codes in the reference mutually exclusive point set, the initial connection mode inside the switch corresponding to the input code is adjusted once to obtain a candidate connection mode; based on the number of each output mutually exclusive pair, each candidate connection mode is adjusted twice to obtain the target connection mode inside all switches. This method adjusts the connection mode once from a local perspective based on the number of output mutually exclusive pairs of each set of input codes, and then uses the number of each output mutually exclusive pair to adjust the connection mode from an overall perspective. That is, the adjustment process starts from both local and global perspectives, and can adjust the connection mode more comprehensively, so that the target connection mode inside all switches is more accurate. At the same time, the operation process for the reference mutually exclusive point set only includes swapping the connection relationship inside the switch, which can ensure that there must be no input mutually exclusive pairs in the two mutually exclusive point sets, thereby simplifying the calculation logic and reducing the cost of the chiplet interface.
[0103] In a specific embodiment, as shown in FIG9 , the network topology and routing optimization design method for chiplet interconnection includes the following steps:
[0104] S701, determining the input code and output code of the data to be exchanged in the coregrain interface of the coregrain according to the address information of the data to be exchanged;
[0105] S702, based on the output mutually exclusive pair relationship, obtaining the number of output mutually exclusive pairs of any group of input codes in the reference to-be-mutually-exclusive point set;
[0106] S703, based on the number of output mutually exclusive pairs of any set of input codes in the reference mutually exclusive point set, adjust the initial connection mode inside the switch corresponding to the input codes to obtain a candidate connection mode;
[0107] S704, based on the number of mutually exclusive pairs of each output, each candidate connection mode is adjusted twice to obtain the target connection mode inside all switches;
[0108] S705, determining first control codes of multiple switches in the output switch group and the input switch group based on the target connection mode;
[0109] S706, determining second control codes of multiple switches in multiple sub-network topologies according to the first control code and the inter-layer connection;
[0110] S707, controlling the interconnection of the data to be exchanged between adjacent core particles through the control code of each switch.
[0111] Next, from an overall perspective, the specific steps of the above control code determination process are introduced. The input mutually exclusive pairs are initialized to two mutually exclusive point sets X and Y, and the elements in each group of input mutually exclusive pairs are traversed in turn to determine whether the output mutually exclusive elements corresponding to the elements in the current mutually exclusive pair are located in the mutually exclusive point set X. Based on the judgment result, the mutually exclusive flag MUTEX represented by 4 bits is obtained. Each bit in MUTEX corresponds to each group of elements in the mutually exclusive point set of 4 groups of input mutually exclusive pairs, where 0 represents an even element and 1 represents an odd element. According to the first-level switch on the left, the elements in the mutually exclusive point set X are connected to the upper output end of the two output ends of the switch, and for the first-level switch on the right, the elements in the mutually exclusive point set X are connected to the upper input end of the two input ends of the switch. The control code of the current outermost two-level switch is generated using the MUTEX mutually exclusive flag, without considering the mutually exclusive point set. Then, the generated control code and inter-level connections route the bits to be replaced to the middle to generate new mutually exclusive pairs and a new outermost layer, and repeat the previous steps to generate the control code of the current outermost switch until the middle-level switch, and then the control codes of multiple switches in the chiplet interconnection network can be obtained. Based on the control codes of multiple switches, the data to be exchanged can be controlled to interconnect between adjacent chiplets.
[0112] FIG10 is a schematic diagram of an 8×8 BENES network architecture, which is based on the traditional 8×8 BENES network architecture, without involving any improvement to the BENES network architecture, and simply improves the core particle interconnection algorithm to generate the control codes of multiple switches in the BENES network architecture. The routing calculation module (BENES_Route_Algo) in the figure is used to calculate the control code of the 2x2 switch in the 8×8 BENES network according to the address information (ADDRESS_IN) of the data to be exchanged. When the control code is 0, the switch is directly connected, and when the control code is 1, the switch is cross-connected. The first mutex flag calculation module (Comp_MUTEX_FLAG_Algo_p0) and the second mutex flag calculation module (Comp_MUTEX_FLAG_Algo_p1) calculate the mutex flag, which is input into the first control code calculation module (Gen_Outermost_Ctrl_Bits_p0) and the second control code calculation module (Gen_Outermost_Ctrl_Bits_p1) to calculate the control codes of the two outermost switches and the two second outermost switches of the 8x8 BENES network. The middle 2x2 switch requires special processing, that is, the control code is calculated by the third control code calculation module (Gen_Middle_Ctrl_Bits_p2). The input mutex pair determination module (New_Output_Address_L1_p1) and the output mutex pair determination module (New_Output_Address_L2_p2) generate new input and output mutex pairs according to the current state of the outermost switch and the inter-stage connection, which is the address information of the new data to be exchanged.
[0113] The above embodiments are all about introducing the algorithm optimization process of the chiplet interconnection process. On the basis of algorithm optimization, the topology of the chiplet interconnection network on the chiplet interface can also be improved to further reduce the area of the chiplet interface and reduce the interconnection cost between different chiplets. In one embodiment, the specific content of the topology improvement process of the chiplet interconnection network includes:
[0114] The core-grain interface of the core-grain includes a core-grain interconnection network. The network topology of the core-grain interconnection network is composed of an input switch group, an output switch group and multiple sub-network topologies. Any sub-network topology includes an input sub-switch group and an output sub-switch group. The input sub-switch group includes multiple adjacent input switches. Intra-layer channels are set between adjacent input switches; the intra-layer channels are used for data interaction between adjacent input switches.
[0115] Taking the dynamic multi-stage blocking network (Butterfly network) with two sub-network topologies of 4×4 in the core-grain interconnection network as an example, Figures 11 and 12 show the data routing diagrams of the sub-network topology with blocking. The bold lines in the figures represent the paths of data transmission, and the black dots represent the blocked ports of the switches, that is, the two inputs of the same switch in the left level need to be exchanged to the two outputs of the same switch in the right level.
[0116] Figures 13 and 14 show the data routing diagram of the non-blocking sub-network topology. By setting an intra-layer channel between the two input switches of the sub-network topology to increase the data path between the input switch and the output switch, blocking can be avoided. Compared with Figures 11 and 12, the core interconnect network with the added intra-layer channel is non-blocking like the traditional non-blocking 4×4 BENES network, but it reduces 2 2-to-1 data selectors and 1-bit control code, which greatly reduces the required hardware resources and control logic. The improved 4x4 Butterfly network can serve as the intermediate level of the 8x8 BENES network, which can reduce the topological area of the network itself and the hardware resource overhead of the control code generation logic.
[0117] FIG15 is a circuit diagram of a conventional 4×4 BENES network, FIG16 is a circuit diagram of a 4×4 Butterfly network, where 0.0 and 0.1 refer to the numbers of switches close to the input end, 1.0 and 1.1 refer to the numbers of switches close to the output end, and FIG17 is a schematic diagram of the output of a 4×4 Butterfly network. The first and second output situations corresponding to the sub-graphs in the upper left and upper right corners can directly fix the first 3 bits of the control code according to the original input, while the output situation corresponding to the sub-graph in the lower left corner is consistent with the conventional blocking 4x4 Butterfly network. Based on this feature, the calculation logic of the control code of the 4x4 Butterfly network can be optimized separately.
[0118] Specifically, in one embodiment, each of the adjacent input switches includes an interlayer interface and an intralayer interface, the interlayer interface is connected to multiple switches of the output sub-switch group; the intralayer interface is used to connect to another switch in the adjacent input switch.
[0119] Continuing to refer to FIG. 13 and FIG. 14 , it can be seen from the figure that each of the adjacent input switches includes four interfaces, two of which are inter-layer interfaces, that is, interfaces connected to two output switches, and the other two interfaces are intra-layer interfaces, which are interfaces for interaction between two input switches.
[0120] In the above-mentioned network topology and routing optimization design method for chiplet interconnection, the chiplet interface of the chiplet includes a chiplet interconnection network, and the network topology of the chiplet interconnection network is composed of an input switch group, an output switch group and multiple sub-network topologies. Any sub-network topology includes an input sub-switch group and an output sub-switch group. The input sub-switch group includes multiple adjacent input switches, and an intra-layer channel is set between adjacent input switches; the intra-layer channel is used for data interaction between adjacent input switches. Each of the adjacent input switches includes an inter-layer interface and an intra-layer interface, and the inter-layer interface is connected to multiple switches of the output sub-switch group; the intra-layer interface is used to connect to another switch in the adjacent input switch. By adding an intra-layer interface to the input switch in the sub-network topology, the number of switches in the sub-network topology can be reduced. At the same time, the process can also reduce the number of control codes. Based on this, the control code generation logic is simplified for the improved sub-network topology, thereby reducing the chiplet interface area overhead and reducing the chiplet interconnection cost.
[0121] FIG18 is a schematic diagram of the improved 8×8 BENES network architecture. It can be seen from the figure that the 8×8 BENES network is composed of two non-blocking 4x4 Butterfly networks, an input switch group, and an output switch group. The change in the topology of the non-blocking 4x4 Butterfly network also brings about the optimization of the control code generation logic. For the 4x4 Butterfly network, it is only necessary to determine whether the current output situation corresponds to a blocking situation of a 4x4 Butterfly network based on the output addresses corresponding to the two inputs on any input switch, and directly determine the control codes of some switches. The remaining switch control codes only need to be generated according to the addressing characteristics of the original Butterfly network with blocking characteristics.
[0122] In the specific design process, the code corresponding to the core particle interaction method is written by the corresponding software, and it is comprehensively known under the 40 nanometer (nm) process that the area overhead of the 8x8 BENES network after only optimizing the core particle interaction method is 78.30% before optimization, and the power consumption is 97.32% before optimization. This shows that the core particle interaction method mentioned in this application can reduce the hardware resource overhead of the switch control code generation logic from an algorithmic perspective. The area overhead of the 8x8 BENES network after combining the core particle interaction method and the network topology is 65.94% before optimization, and the power consumption is 93.78% before optimization. Therefore, optimizing the network topology also effectively reduces the hardware resource overhead of the 8x8 BENES network. It can be seen that through the optimization of the algorithm and the optimization of the network topology of the core particle interaction method, the area of the core particle interface can be greatly reduced, thereby reducing the interconnection cost between different core particles.
[0123] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0124] Based on the same inventive concept, the embodiment of the present application also provides a network topology and routing optimization design device for chiplet interconnection, which is used to implement the network topology and routing optimization design method for chiplet interconnection. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the network topology and routing optimization design device for chiplet interconnection provided below can refer to the limitations of the network topology and routing optimization design method for chiplet interconnection above, and will not be repeated here.
[0125] In an exemplary embodiment, as shown in FIG19 , a network topology and routing optimization design device for chiplet interconnection is provided, including: a coding determination module 11, a control code determination module 12 and a control module 13, wherein:
[0126] The coding determination module 11 is used to determine the input coding and output coding of the data to be exchanged in the chiplet interface of the chiplet according to the address information of the data to be exchanged;
[0127] A control code determination module 12, configured to determine control codes of a plurality of switches of a chiplet interconnection network in a chiplet interface based on a preset output mutually exclusive pair relationship, an input code, and an output code;
[0128] The control module 13 is used to control the interconnection of the data to be exchanged between adjacent core particles through the control code of each switch.
[0129] In one embodiment, the control code determination module includes: an acquisition unit and a control code determination unit, wherein:
[0130] An acquisition unit, configured to acquire first control codes of a plurality of switches in an input switch group and an output switch group based on a preset output mutually exclusive pair relationship, an input code, and an output code;
[0131] The control code determination unit is used to determine the second control codes of multiple switches in multiple sub-network topologies according to the first control code and the inter-layer connection.
[0132] In one embodiment, the acquisition unit is also used to acquire the target connection mode inside all switches in the output switch group and the input switch group based on the output mutually exclusive pair relationship, the input code and the output code; based on the target connection mode, determine the first control code of multiple switches in the output switch group and the input switch group.
[0133] In one embodiment, the acquisition unit is also used to acquire the number of output mutually exclusive pairs of any group of input codes in the reference set of points to be mutually exclusive based on the output mutually exclusive pair relationship; based on the number of output mutually exclusive pairs, determine the target connection mode inside all switches in the input switch group and the output switch group.
[0134] In one embodiment, the acquisition unit is further used to adjust the initial connection mode inside the switch corresponding to the input code based on the number of output mutually exclusive pairs of any group of input codes in the reference set of mutually exclusive points to obtain a candidate connection mode; and to adjust each candidate connection mode a second time based on the number of each output mutually exclusive pair to obtain a target connection mode inside all switches.
[0135] In one embodiment, the chiplet interface of the chiplet includes a chiplet interconnection network, the network topology of the chiplet interconnection network is composed of an input switch group, an output switch group and multiple sub-network topologies, any sub-network topology includes an input sub-switch group and an output sub-switch group, the input sub-switch group includes multiple adjacent input switches, and intra-layer channels are set between adjacent input switches; the intra-layer channels are used for data interaction between adjacent input switches.
[0136] In one embodiment, each of the adjacent input switches includes an interlayer interface and an intralayer interface, wherein the interlayer interface is connected to multiple switches of the output sub-switch group; and the intralayer interface is used to connect to another switch in the adjacent input switches.
[0137] Each module in the above-mentioned network topology and routing optimization design device for chiplet interconnection can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.
[0138] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the content of any one of the embodiments of the network topology and routing optimization design method for chiplet interconnection is implemented.
[0139] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the content of any embodiment of the network topology and routing optimization design method for chiplet interconnection is implemented.
[0140] In one embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements the content of any one of the embodiments of the network topology and routing optimization design method for chiplet interconnection.
[0141] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0142] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.
[0143] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.
[0144] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. A network topology and routing optimization design method for chiplet interconnection, characterized in that: The method is applied to core particles, and the method comprises: Determine the input code and output code of the data to be exchanged in the coregrain interface of the coregrain according to the address information of the data to be exchanged; Determining control codes of a plurality of switches of a chiplet interconnection network in the chiplet interface based on a preset output mutually exclusive pair relationship, the input code, and the output code; The control code of each switch is used to control the data to be exchanged to be interconnected between adjacent core particles.
2. The method according to claim 1, characterized in that The core-grain interconnection network includes an input switch group, an output switch group, and a plurality of sub-network topologies, wherein the plurality of sub-network topologies are connected to the input switch group and the output switch group through inter-layer connections; The step of determining the control codes of the plurality of switches of the chiplet interconnection network in the chiplet interface based on the preset output mutually exclusive pair relationship, the input code and the output code comprises: Based on a preset output mutually exclusive pair relationship, the input code and the output code, obtaining first control codes of a plurality of switches in the input switch group and the output switch group; According to the first control code and the inter-level connections, second control codes of multiple switches in the multiple sub-network topologies are determined.
3. The method according to claim 2, characterized in that The acquiring, based on the preset output mutually exclusive pair relationship, the input code and the output code, the first control code of the plurality of switches in the input switch group and the output switch group comprises: Based on the output mutually exclusive pair relationship, the input code and the output code, obtaining the target connection mode inside all switches in the output switch group and the input switch group; Based on the target connection mode, first control codes of a plurality of switches in the output switch group and the input switch group are determined.
4. The method according to claim 3, characterized in that The two input codes of each input switch in the input switch group are an input mutually exclusive pair; The acquiring the target connection mode inside all switches in the input switch group and the output switch group based on the output mutually exclusive pair relationship, the input code and the output code comprises: Based on the output mutually exclusive pair relationship, obtaining the number of output mutually exclusive pairs of any group of input codes in the reference set of points to be mutually exclusive; Based on the number of the output mutually exclusive pairs, a target connection mode inside all switches in the input switch group and the output switch group is determined.
5. The method according to claim 4, characterized in that The determining, based on the number of output mutually exclusive pairs, the target connection mode inside all switches in the input switch group and the output switch group includes: Based on the number of output mutually exclusive pairs of the arbitrary set of input codes in the reference set of mutually exclusive points, an initial connection mode inside the switch corresponding to the input codes is adjusted to obtain a candidate connection mode; Based on the number of each output mutually exclusive pair, each candidate connection mode is adjusted twice to obtain the target connection mode inside all switches.
6. The method according to any one of claims 1 to 5, characterized in that: The core-grain interface of the core-grain includes a core-grain interconnection network, the network topology of the core-grain interconnection network is composed of an input switch group, an output switch group and multiple sub-network topologies, any sub-network topology includes an input sub-switch group and an output sub-switch group, the input sub-switch group includes multiple groups of adjacent input switches, and an intra-layer channel is set between each group of adjacent input switches; the intra-layer channel is used for data interaction between the adjacent input switches.
7. The method according to claim 6, characterized in that Each of the adjacent input switches includes an inter-layer interface and an intra-layer interface, wherein the inter-layer interface is connected to a plurality of switches of the output sub-switch group; and the intra-layer interface is used to connect to another switch of the adjacent input switches.
8. A network topology and routing optimization design device for chiplet interconnection, characterized in that: The device comprises: A coding determination module, used for determining the input coding and output coding of the data to be exchanged in the coregrain interface of the coregrain according to the address information of the data to be exchanged; A control code determination module, configured to determine control codes of a plurality of switches of a chiplet interconnection network in the chiplet interface based on a preset output mutually exclusive pair relationship, the input code and the output code; The control module is used to control the interconnection of the data to be exchanged between adjacent core particles through the control code of each switch.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.
Citation Information
Patent Citations
Port remapping circuit and method for network switching chip
CN108462660A
Method for implementing photoelectric mutex, ethernet photoelectric mutex interface device and network equipment
US20090323705A1
Multi-media serial line switching adapter for parallel networks and heterogeneous and homologous computer system
US5774698A
Cited By
Chip routing method and device, electronic equipment and storage medium
CN121349950A
