A data transfer module, an integrated chip, and a vertically scalable network architecture

By introducing switching chips and non-fiber optical guide components into the optical interconnection between computing chips, the limitations of single-chip computing power and the bottleneck of electrical interconnection are solved, realizing efficient multi-chip collaboration and low-latency communication, thus meeting the needs of high-performance computing.

CN122673148BActive Publication Date: 2026-11-17HANGZHOU KAIKAI TECHNOLOGY CO LTD +1
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611164291.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-08-03
Publication Date
2026-11-17
Estimated Expiration
2046-08-03

AI Technical Summary

Technical Problem

Existing computing chips are limited by single-chip computing power and electrical interconnect bottlenecks in high-performance computing scenarios, making it difficult to meet the explosive growth in computing power demand. Furthermore, traditional fiber optic interconnects present alignment challenges and high packaging costs.

Method used

By employing a data transfer module and a switching chip, an optical transmitting unit converts electrical signals into modulated laser beams. Optical interconnection is achieved through multi-level mirror reflection in free space using a non-fiber optical guide component. Combined with a receiving unit to restore the electrical signals, a highly efficient multi-chip collaborative system is constructed.

Benefits of technology

Breaking through the physical limits of single-chip communication, it achieves low-latency, high-bandwidth inter-chip communication, reduces signal loss and latency, simplifies packaging processes, and improves computing performance and integration.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122673148B_ABST
    Figure CN122673148B_ABST
Patent Text Reader

Abstract

This application provides a data transfer module, an integrated chip, and a vertically scalable network architecture, aiming to overcome the computing power bottleneck of a single computing chip. The data transfer module includes an optical transmitting unit, an optical receiving unit, and a non-fiber optical guide component located on the same side of the carrier. It is electrically connected to the computing or storage chip via a switching chip and utilizes at least two stages of specular reflection to achieve optical path transmission. The integrated chip integrates the above module, the switching chip, and other chips onto the same carrier, achieving high-density computing power aggregation and efficient optical interconnection. The vertically scalable network architecture uses multiple integrated chips to form a computing power cluster. This application replaces traditional electrical interconnection with optical interconnection, significantly reducing signal latency and power consumption, improving bandwidth and integration, and providing a feasible solution for high-performance computing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of computing chip technology, and in particular to a data transfer module, an integrated chip, and a vertically scalable network architecture. Background Technology

[0002] With the rapid development of technologies such as artificial intelligence, big data analytics, and high-performance computing, the demand for computing power in computing chips has exploded. In applications such as AI training, scientific computing simulation, and real-time data processing, computing chips need to complete massive data processing tasks in a very short time, which places extremely high demands on the chip's computing speed and data processing capabilities.

[0003] Currently, the industry primarily meets computing power demands by improving the manufacturing process of individual computing chips, increasing the number of cores, or optimizing the architecture design. However, as Moore's Law gradually slows down, chip manufacturing processes are approaching their physical limits, and further improvements in integration and computing speed face multiple technical bottlenecks, including excessive power density, heat dissipation difficulties, and signal latency. Simply relying on upgrades to single chips is no longer sufficient to meet the ever-increasing computing power demands; the physical limitations of chip computing power have become a key factor restricting overall computing performance improvements. Summary of the Invention

[0004] The purpose of this application is to provide a data transfer module, an integrated chip, and a vertically scalable network architecture that can solve the above-mentioned problems.

[0005] The embodiments of this application are implemented as follows: In a first aspect, this application discloses a data transfer module, including a light emitting unit, a light receiving unit, and a non-fiber light guiding component disposed on the same side of the carrier; When the optical receiving unit is electrically connected to the switching chip, the optical transmitting unit is electrically connected to other chips that generate and / or store data; or, when the optical transmitting unit is electrically connected to the switching chip, the optical receiving unit is electrically connected to other chips that generate and / or store data. The non-fiber optical guide component is disposed on the surface of the carrier and located between the other chips and the switching chip. It is configured to guide the target modulated laser beam output by the optical emitting unit into the optical receiving unit through at least two levels of specular reflection in free space.

[0006] It is understood that the data transfer module disclosed in the first aspect of this application effectively solves the problems of limited computing power of a single chip and bottlenecks in electrical interconnection in the prior art by introducing a switching chip between at least two computing chips. In this module, the switching chip receives target data from one computing chip through the data transfer module, processes the data according to a preset protocol (such as data routing and switching control), and accurately forwards it to another computing chip according to a preset path. The presence of the switching chip enables efficient collaboration and data aggregation between multiple chips, allowing multiple computing chips to work as a logical whole, thereby breaking through the physical limits of a single chip and realizing the vertical expansion of computing power. At the same time, the data transfer module converts electrical signals into modulated laser beams through an optical transmitting unit, emits them toward the carrier, and then guides them to an optical receiving unit through a non-fiber optical guide component, and then restores them to electrical signals. This optical interconnection method avoids the signal attenuation, delay, and power consumption problems of traditional electrical interconnection in high-frequency transmission, and the data transfer module and the chip are set on the same side of the same carrier, resulting in a compact structure and high integration. In summary, this module achieves low-latency, high-bandwidth inter-chip communication through intelligent scheduling of the switching chip and high-speed transmission of optical interconnects, providing a feasible solution for building high-performance computing clusters.

[0007] In addition, the data transfer module uses a non-fiber optical fiber light guide component for beam guidance, instead of traditional optical fiber. This design has significant advantages: it avoids the alignment difficulties and complex processes of coupling optical fiber to the chip, reducing packaging difficulty and cost; at the same time, the non-fiber optical fiber light guide component can be directly integrated onto the carrier, supporting a more compact chip layout, shortening the optical path transmission distance, and further reducing signal loss and latency.

[0008] In an optional embodiment of this application, the non-fiber light guiding component includes a groove disposed on a carrier or other layer structure on the carrier, the groove including at least two reflective surfaces, the at least two reflective surfaces guiding the target modulated laser beam output by the light emitting unit into the light receiving unit.

[0009] In an optional embodiment of this application, the groove includes a first inner sidewall and a second inner sidewall disposed opposite to each other. A first reflective surface is disposed on the first inner sidewall, and a second reflective surface is disposed on the second inner sidewall. The first reflective surface is configured to reflect the target modulated laser beam output by the light emitting unit, causing it to exit in a direction parallel to the carrier surface. The second reflective surface is configured to reflect the target modulated laser beam incident in a direction parallel to the carrier surface, causing it to be guided into the light receiving unit.

[0010] It is understandable that by directly etching a groove with a reflective surface onto the carrier as a non-fiber light guide component, monolithic integration of the optical path structure and the carrier is achieved. This design eliminates the need for assembling independent light guide elements, greatly simplifies the packaging process, improves alignment accuracy and structural stability, while shortening the optical path transmission distance, effectively reducing signal loss and delay, and lowering costs.

[0011] In an optional embodiment of this application, the data transfer module includes a receiving data transfer module and a transmitting data transfer module; the optical receiving unit in the receiving data transfer module is electrically connected to the switching chip, and the optical transmitting unit in the receiving data transfer module is electrically connected to other chips that generate and / or store data; the optical transmitting unit in the transmitting data transfer module is electrically connected to the switching chip, and the optical receiving unit in the transmitting data transfer module is electrically connected to other chips that generate and / or store data.

[0012] It is understandable that by distinguishing between receiving and transmitting data transfer modules, bidirectional and orderly optical communication between the switching chip and other chips is achieved. The receiving module is responsible for transmitting data from other chips to the switching chip, while the transmitting module performs the reverse transmission. This clearly defined modular design simplifies the optical path layout and control logic, and improves the flexibility and efficiency of bidirectional data transmission in the system.

[0013] In an optional embodiment of this application, the other chip includes at least one of the following: a computing chip or a storage chip.

[0014] In this embodiment, the other chips can be computing chips or storage chips, specifically including the following situations: First, the other chips are computing chips. Examples include central processing units (CPUs), graphics processing units (GPUs), neural network processors (NPUs), tensor processors (TPUs), field-programmable gate arrays (FPGAs), or application-specific integrated circuits (ASICs). In this case, the data transfer module is used to establish a high-speed optical communication link between the switching chip and the computing chip, enabling data collaboration and aggregation among multiple computing chips. This allows multiple computing chips to work in parallel as a logical whole, breaking through the single-chip computing power limit and meeting the needs of high-performance scenarios such as AI training and scientific computing. Second, the other chips are storage chips. Examples include high-bandwidth memory (HBM), dynamic random access memory (DRAM), static random access memory (SRAM), or non-volatile memory (NVM). In this case, the optical interconnect module is used for high-speed data access between the switching chip and the storage chip, solving the bottlenecks of traditional electrical interconnects in terms of storage bandwidth, access latency, and power consumption, and improving data throughput efficiency. Third, the other chips simultaneously include computing chips and storage chips. Multiple computing chips and storage chips can be heterogeneously integrated through switching chips and optical interconnect modules. The switching chip can be connected to both chips through optical interconnect modules to achieve flexible scheduling and efficient collaboration of computing and storage resources, and build a high-density, low-latency computing and storage unit.

[0015] In an optional embodiment of this application, the light emitting unit includes a transmitting end support plate and a laser chip and a driving chip disposed on the same side of the transmitting end support plate; the driving chip is electrically connected to the switching chip or the other chip, and a microcircuit is disposed on the surface of the transmitting end support plate, the microcircuit being configured to electrically connect the laser chip and the driving chip; the light emitting surface of the laser chip faces the carrier and is configured to emit a target modulated laser beam toward the carrier.

[0016] In optional embodiments of this application, the driver chip is electrically connected to the switching chip or the other chip, including at least one of the following: 1. The driver chip is directly bonded to the switching chip or other chips to achieve electrical connection. It can be understood that this embodiment eliminates intermediate connection structures by directly bonding the driver chip to the switching chip or other chips, greatly shortening the electrical signal transmission path. This effectively reduces signal attenuation and parasitic parameters, improves transmission rate and integrity, simplifies packaging processes, and enhances system integration and reliability.

[0017] 2. An intermediate adapter board is provided between the driver chip and the switching chip or other chips. The driver chip is bonded to the surface of the intermediate adapter board facing the transmitting end support plate, and the surface of the intermediate adapter board facing away from the transmitting end support plate is bonded to the switching chip or other chips. The intermediate adapter board is provided with conductive vias to connect corresponding solder joints on the two surfaces, thereby realizing the electrical connection between the driver chip and the switching chip or other chips. It can be understood that this embodiment achieves heterogeneous integration bonding between the driver chip and the switching chip or other chips through an intermediate adapter board. This structure utilizes the conductive vias on the adapter board to achieve vertical interconnection, effectively solving the problem of spacing or package mismatch that may exist between different chips while ensuring electrical connection performance, thus improving design flexibility and integration density.

[0018] In an optional embodiment of this application, the optical receiving unit includes a receiving end support plate and a receiving processing chip and a photodetector chip disposed on the same side of the receiving end support plate; the photodetector chip is configured to receive the target modulated laser beam, and a microcircuit is disposed on the surface of the receiving end support plate, the microcircuit being configured to electrically connect the photodetector chip and the receiving processing chip; the receiving processing chip is electrically connected to the other chips or the switching chip.

[0019] In optional embodiments of this application, the receiving processing chip is electrically connected to the other chips or the switching chip, including at least one of the following: 1. The receiving and processing chip is directly bonded to the other chips or the switching chip to achieve electrical connection. It can be understood that this embodiment eliminates intermediate connection structures by directly bonding the receiving and processing chip to other chips or the switching chip, greatly shortening the electrical signal transmission path. This effectively reduces signal attenuation and parasitic effects, improves data transmission rate and integrity, simplifies packaging processes, and enhances system integration and reliability.

[0020] 2. An intermediate adapter board is provided between the receiving and processing chip and the other chips or the switching chip. The receiving and processing chip is bonded to the surface of the intermediate adapter board facing the receiving end support plate, and the surface of the intermediate adapter board facing away from the receiving end support plate is bonded to the other chips or the switching chip. The intermediate adapter board is provided with conductive vias to connect corresponding solder joints on the two surfaces, thereby realizing the electrical connection between the receiving and processing chip and the other chips or the switching chip. It can be understood that this embodiment achieves heterogeneous integration of the receiving and processing chip with other chips or the switching chip through the intermediate adapter board. Utilizing the conductive vias on the adapter board to achieve vertical interconnection effectively solves the problem of chip spacing or package mismatch while ensuring high-speed signal transmission quality, improving design flexibility and integration density.

[0021] Secondly, this application discloses an integrated chip, including a carrier and at least one data transfer module as described in any of the first aspects, at least one switching chip, and at least one other chip that generates and / or stores data, disposed on the same side of the carrier.

[0022] It is understandable that integrating multiple computing chips, switching chips, and data transfer modules onto the same carrier and side creates a high-density computing unit. By replacing traditional electrical interconnects with optical interconnects, bandwidth, power consumption, and latency bottlenecks are overcome, effectively aggregating multi-chip computing power, surpassing the physical limits of a single chip, significantly improving inter-chip data exchange efficiency, and providing core support for building high-performance computing clusters.

[0023] In an optional embodiment of this application, the other chips include a memory chip; the switching chip is electrically connected to the memory chip through the data transfer module; when the optical receiving unit of the data transfer module is electrically connected to the switching chip and the optical transmitting unit is electrically connected to the memory chip, the switching chip reads target data from the memory chip; when the optical transmitting unit of the data transfer module is electrically connected to the switching chip and the optical receiving unit is electrically connected to the memory chip, the switching chip stores the target data into the memory chip.

[0024] It is understandable that optical interconnect modules enable bidirectional high-speed data access between switching chips and storage chips. The switching chip can both read and write data to the storage chip, overcoming the bottlenecks of traditional electrical interconnects in terms of storage bandwidth, access latency, and power consumption. This significantly improves data throughput efficiency, achieves efficient collaboration between computing and storage resources, and provides support for the construction of high-density, low-latency compute-memory integrated units.

[0025] In an optional embodiment of this application, the integrated chip includes a first receiving data transfer module and a first transmitting data transfer module; the switching chip is configured to be electrically connected to the first receiving data transfer module and the first transmitting data transfer module respectively; one end of the first receiving data transfer module opposite to the switching chip is configured to be electrically connected to a first computing chip, and one end of the first transmitting data transfer module opposite to the switching chip is configured to be electrically connected to a second computing chip.

[0026] It is understandable that the first receiver and the first transmitter optical interconnect modules enable efficient unidirectional communication between the switching chip and two different computing chips. The switching chip can acquire data from the first computing chip via the receiver module and simultaneously send data to the second computing chip via the transmitter module, forming a clear pipeline path. This design simplifies optical path layout and signal scheduling, and improves the efficiency of multi-chip collaborative processing.

[0027] In an optional embodiment of this application, the integrated chip further includes a second receiving data transfer module and a second transmitting data transfer module electrically connected to the switching chip; one end of the second receiving data transfer module opposite to the switching chip is configured to be electrically connected to the second computing chip, and the other end of the second transmitting data transfer module opposite to the switching chip is configured to be electrically connected to the first computing chip.

[0028] It is understandable that by adding a second receiver and a second transmitter module to the first receiver and first transmitter modules, full-duplex bidirectional communication between the switching chip and the first and second computing chips is achieved. The switching chip can simultaneously receive data from and send data to both chips, forming a symmetrical interconnect structure. This design significantly improves data exchange bandwidth and parallel processing capabilities, enhancing the system's communication flexibility.

[0029] In optional embodiments of this application, the integrated chip includes at least one of the following: The number of the first receiving data transfer modules is greater than or equal to 2; The number of the first transmitting data transfer modules is greater than or equal to 2; The number of the second receiving data transfer modules is greater than or equal to 2; The number of the second transmitting data transfer modules is greater than or equal to 2.

[0030] It is understandable that by configuring multiple first or second receiver-transmitter optical interconnect modules, parallel extended communication between the switching chip and multiple computing chips is achieved. The multi-channel design significantly improves the total bandwidth and data throughput, supporting more complex multi-chip collaborative computing scenarios. At the same time, the flexible configuration of the number of modules enhances the system's scalability, which can be dynamically adjusted according to computing power requirements to meet the large-scale deployment requirements of high-performance clusters.

[0031] In an optional embodiment of this application, the switching chip includes a first side and a second side disposed opposite to each other, the first computing chip being disposed near the first side and the second computing chip being disposed near the second side.

[0032] It is understandable that placing the first computing chip near the first side of the switching chip and the second computing chip near the opposite second side creates a symmetrical and separated physical layout. This arrangement effectively shortens the transmission distance of the optical interconnect module and reduces signal delay and loss. At the same time, the opposite side layout avoids signal cross-interference, simplifies optical path design and wiring complexity, and helps improve the communication efficiency and integration reliability of multi-chip systems.

[0033] In an optional embodiment of this application, the switching chip includes a first side and a third side disposed adjacent to each other, the first computing chip is disposed near the first side, and the second computing chip is disposed near the third side.

[0034] It is understandable that placing the first computing chip near the first side of the switching chip and the second computing chip near the adjacent third side creates a compact L-shaped layout. This arrangement allows for the integration of more chips within a limited space, increasing packaging density. Simultaneously, the adjacent side layout shortens the optical path corner distance, reduces reflection loss and signal delay, and optimizes space utilization and communication efficiency.

[0035] In an optional embodiment of this application, the switching chip and the other chips are all bare chips, the carrier is a circuit board, the bare chips are electrically connected to the surface of the circuit board, and the non-fiber optical guide component is disposed in a groove on the circuit board.

[0036] It is understandable that directly connecting the bare chip to the circuit board surface and integrating the non-fiber optical guide component into a recess in the circuit board achieves high-density integration of the chip and the optical path on the same substrate. This design eliminates intermediate transition structures, simplifies packaging layers, shortens the electrical signal transmission path, and reduces parasitic parameters and signal loss. At the same time, the compact layout of the recessed optical guide component and the bare chip improves the overall integration and communication reliability.

[0037] In an optional embodiment of this application, the switching chip and the other chips are all bare chips, the carrier is a carrier substrate, and the carrier substrate is provided with at least one light guide carrier plate and at least two circuit boards; each bare chip is electrically connected to a single circuit board, and each non-fiber light guide component is fixed to a single light guide carrier plate, wherein the non-fiber light guide component is a groove provided on the light guide carrier plate.

[0038] It is understood that this embodiment integrates the light guide carrier plate and the circuit board through a carrier substrate, realizing the modular and partitioned assembly of the optical path and the circuit. The bare core is independently electrically connected to the circuit board, and the non-fiber light guide components are independently fixed to the light guide carrier plate, which reduces the difficulty of alignment and coupling, improves the assembly yield and maintainability, and facilitates the separate optimization of the optical path and the circuit.

[0039] In optional embodiments of this application, the other chip is a calculator circuit board carrying a bare computing chip and its peripheral auxiliary circuits; the switching chip is a switch circuit board carrying a bare switching chip and its peripheral auxiliary circuits; the carrier is a substrate, and the calculator circuit board and the switch circuit board are fixed on the substrate; the non-fiber optic light guide component is a groove disposed on the substrate.

[0040] In optional embodiments of this application, the other chip is a memory circuit board carrying a bare memory chip and its peripheral auxiliary circuits; the switching chip is a switch circuit board carrying a bare switching chip and its peripheral auxiliary circuits; the carrier is a substrate, and the memory circuit board and the switch circuit board are fixed on the substrate; the non-fiber optic light guide component is a groove disposed on the substrate.

[0041] In optional embodiments of this application, the peripheral auxiliary circuit includes at least one of the following: a power management circuit, a clock circuit, a data interface circuit, and a storage circuit.

[0042] It is understandable that the computing or storage bare chip and its peripheral auxiliary circuitry are integrated onto a separate circuit board, then fixed to a substrate, and optical interconnects are achieved using substrate recesses. This modular design facilitates independent testing, replacement, and maintenance of each component, reducing alignment difficulties and packaging costs. Simultaneously, the proximity of peripheral circuitry optimizes power integrity and signal quality, improves system reliability and integration flexibility, and is beneficial for mass production.

[0043] Thirdly, this application discloses a vertically scalable network architecture, which employs a computing power cluster composed of multiple integrated chips as described in any one of the second aspects.

[0044] The vertically scalable network architecture disclosed in the third aspect of this application can be a scale-up architecture. A scale-up architecture refers to improving system performance by enhancing the computing power of individual nodes (such as increasing CPU / memory or integrating more chips), rather than increasing the number of nodes. The scale-up architecture in this application uses inter-chip optical interconnect technology to interconnect multiple integrated chips at high speed, aggregating computing power and enabling them to work collaboratively like a powerful "single computer" to meet the demands of high-performance computing. This architecture, by employing multiple of the aforementioned integrated chips to build a high-density computing cluster, aims to meet the core requirements of computing power aggregation and efficient collaboration in large-scale parallel computing scenarios.

[0045] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, optional embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a top view of the first integrated chip provided in this application.

[0048] Figure 2 yes Figure 1 The first dashed cross-sectional view of the integrated chip shown.

[0049] Figure 3 yes Figure 1 The second dashed cross-sectional view of the integrated chip shown.

[0050] Figure 4 yes Figure 1 The third dashed cross-sectional view of the integrated chip shown.

[0051] Figure 5 yes Figure 1 The fourth dashed cross-sectional view of the integrated chip shown.

[0052] Figure 6 This is a top view of the second type of integrated chip provided in this application.

[0053] Figure 7 This is a top view of the third type of integrated chip provided in this application.

[0054] Figure 8 This is a top view of the fourth type of integrated chip provided in this application. Detailed Implementation

[0055] 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 scope of protection of this application.

[0056] In a first aspect, this application discloses a data transfer module, including an optical emitting unit, an optical receiving unit, and a non-fiber optic light guide component disposed on the same side of a carrier; when the optical receiving unit is electrically connected to a switching chip, the optical emitting unit is electrically connected to other chips that generate and / or store data; or, when the optical emitting unit is electrically connected to the switching chip, the optical receiving unit is electrically connected to other chips that generate and / or store data; the non-fiber optic light guide component is disposed on the surface of the carrier and located between the other chips and the switching chip, and is configured to guide the target modulated laser beam output by the optical emitting unit into the optical receiving unit through at least two levels of specular reflection in free space.

[0057] In an optional embodiment of this application, the other chip includes at least one of the following: a computing chip or a storage chip.

[0058] In this embodiment, the other chips can be computing chips or storage chips, specifically including the following situations: First, the other chips are computing chips. Examples include central processing units (CPUs), graphics processing units (GPUs), neural network processors (NPUs), tensor processors (TPUs), field-programmable gate arrays (FPGAs), or application-specific integrated circuits (ASICs). In this case, the data transfer module is used to establish a high-speed optical communication link between the switching chip and the computing chip, enabling data collaboration and aggregation among multiple computing chips. This allows multiple computing chips to work in parallel as a logical whole, breaking through the single-chip computing power limit and meeting the needs of high-performance scenarios such as AI training and scientific computing. Second, the other chips are storage chips. Examples include high-bandwidth memory (HBM), dynamic random access memory (DRAM), static random access memory (SRAM), or non-volatile memory (NVM). In this case, the optical interconnect module is used for high-speed data access between the switching chip and the storage chip, solving the bottlenecks of traditional electrical interconnects in terms of storage bandwidth, access latency, and power consumption, and improving data throughput efficiency. Third, the other chips simultaneously include computing chips and storage chips. Multiple computing chips and storage chips can be heterogeneously integrated through switching chips and optical interconnect modules. The switching chip can be connected to both chips through optical interconnect modules to achieve flexible scheduling and efficient collaboration of computing and storage resources, and build a high-density, low-latency computing and storage unit.

[0059] In optional embodiments of this application, the aforementioned computing chip can be selected from any one of a Graphics Processing Unit (GPU), a custom AI accelerator XPU, or an Application-Specific Integrated Circuit (ASIC) chip. The custom AI accelerator XPU is not a single chip, but rather a strategic concept for heterogeneous computing, designed to address complex computing needs through diverse hardware combinations. Its essence is "Right Tool for the Right Job," meaning using the most suitable hardware to handle specific tasks, thereby improving overall system efficiency. The custom AI accelerator XPU can be, for example, a Central Processing Unit (CPU), a Neural Processing Unit (NPU), a Field-Programmable Gate Array (FPGA), or at least one of an AI accelerator. For example, the XPU can be any one of a CPU, GPU, or NPU to accelerate the training of large-scale models. Application-Specific Integrated Circuit (ASIC) chips include serializer / deserializer (SerDes) chips used to encode and decode data through a Digital Signal Processor (DSP) interface.

[0060] like Figure 1 As shown, the switching chip 3 is disposed between the first computing chip 1 and the second computing chip 2, the first data transfer module 10 is disposed between the switching chip 3 and the first computing chip 1, and the second data transfer module 20 is disposed between the switching chip 3 and the second computing chip 2. The first data transfer module 10 is a receiving data transfer module, and the second data transfer module 20 is a transmitting data transfer module. The receiving and transmitting functions are relative to the switching chip 3.

[0061] The switching chip is configured to receive target data from a computing chip through at least one data transfer module, process the target data according to a preset protocol, and then send it to another computing chip through at least one data transfer module according to a preset path.

[0062] like Figure 1 The switching chip 3 is configured to receive target data from the first computing chip 1 through the first data transfer module 10, process the target data according to a preset protocol, and then transmit the processed target data to the second computing chip 2 through the second data transfer module 20 according to a preset path, thereby realizing the data transfer between the first computing chip 1 and the second computing chip 2.

[0063] The above-mentioned processing of target data according to a preset protocol includes protocol conversion or data reassembly of the target data; the preset path includes a preset routing strategy or target address information to determine the transmission path.

[0064] Switching chip 3 is the core of the inter-chip optical interconnect. It is responsible for receiving data from one computing chip, performing protocol conversion or data reassembly, and accurately forwarding it to another computing chip according to a preset routing strategy. It enables efficient collaboration and data aggregation among multiple chips, thereby breaking through the physical limitations of a single chip and serving as a key hub for building high-performance computing clusters.

[0065] Each data transfer module includes an optical emitting unit, an optical receiving unit, and a non-fiber optical guide component, all located on the same side of the carrier. The optical emitting unit is configured to convert the received target data into a target modulated laser beam carrying the target optical signal and emit the target modulated laser beam toward the carrier. The optical receiving unit is configured to receive the target modulated laser beam and restore the target optical signal carried by the target modulated laser beam back into the target data.

[0066] The following explanation will use a computing chip as an example.

[0067] In an optional embodiment of this application, the first data transfer module 10 is a receiving data transfer module. For example... Figure 2 As shown, the first data transfer module 10 includes a first light emitting unit 11, a first light receiving unit 12, and a first non-fiber light guiding component 13, all disposed on the same side of the carrier (circuit board 500 in the figure). The first light emitting unit 11 includes a first transmitting end support plate 111 and a first laser chip 112 and a first driving chip 113 disposed on the same side of the first transmitting end support plate 111. The first driving chip 113 is electrically connected to the first computing chip 1. A microcircuit is disposed on the surface of the first transmitting end support plate 111, and the microcircuit is configured to electrically connect the first laser chip 112 and the first driving chip 113. The light emitting surface of the first laser chip 112 faces the circuit board 500 and is configured to emit a target modulated laser beam toward the circuit board 500.

[0068] In optional embodiments of this application, such as Figure 2 As shown, the first optical receiving unit 12 includes a first receiving end support plate 121 and a first receiving processing chip 123 and a first photoelectric detection chip 122 disposed on the same side of the first receiving end support plate 121; the first photoelectric detection chip 122 is configured to receive a target modulated laser beam, and a microcircuit is disposed on the surface of the first receiving end support plate 121, which is configured to electrically connect the first photoelectric detection chip 122 and the first receiving processing chip 123; the first receiving processing chip 123 is electrically connected to the switching chip.

[0069] The aforementioned laser chip uses a vertical-cavity surface-emitting laser (VCSEL), which has the following advantages: it can emit a beam vertically from the surface of a semiconductor wafer and efficiently couple with the first reflecting surface of a non-fiber optical guide component; it has low threshold current and high-speed modulation characteristics, making it suitable for chip-level short-distance optical interconnects; and it is easy to integrate into an array, supporting high-density parallel optical transmission.

[0070] The aforementioned starting support board serves as an integrated platform, with microcircuits on its surface for high-speed signal transmission. The driver chip receives target data from the first computing chip, converts it into corresponding drive signals, and drives the laser chip through the adapter board's microcircuits. This architecture shortens the interconnection distance between the driver chip and the laser chip, ensuring signal integrity and low latency.

[0071] The aforementioned photoelectric detection chip is a key component that converts received optical signals into a weak current. It typically employs a high-speed PIN photodiode or an avalanche photodiode (APD), which possesses high responsivity and low dark current characteristics, enabling it to sensitively capture high-speed modulated optical signals and complete the photoelectric conversion.

[0072] The aforementioned receiving and processing chip includes a transimpedance amplifier (TIA) and a clock data recovery circuit (CDR); the transimpedance amplifier is configured to receive the target electrical signal, convert it into a voltage signal, and amplify it; the clock data recovery circuit is configured to extract the clock signal and shape the data from the amplified voltage signal to restore the target data.

[0073] In optional embodiments of this application, the first driver chip 113 is electrically connected to the first computing chip 1, including at least one of the following: 1. For example Figure 2 As shown, the first driver chip 113 is directly bonded to the first computing chip 1, achieving electrical connection. It can be understood that this embodiment eliminates intermediate connection structures by directly bonding the first driver chip 113 to the computing chip, greatly shortening the electrical signal transmission path. This effectively reduces signal attenuation and parasitic parameters, improves transmission rate and integrity, simplifies packaging processes, and enhances system integration and reliability.

[0074] 2. For example Figure 3As shown, a first intermediate adapter plate 201 is disposed between the first driver chip 113 and the first computing chip 1. The first driver chip 113 is bonded to the surface of the first intermediate adapter plate 201 facing the starting support plate, and the surface of the first intermediate adapter plate 201 facing away from the starting support plate is bonded to the computing chip. Conductive vias are provided on the first intermediate adapter plate 201 to connect corresponding solder points on the two surfaces, thereby realizing the electrical connection between the first driver chip 113 and the computing chip. It can be understood that this embodiment realizes the heterogeneous integration bonding of the first driver chip 113 and the computing chip through the first intermediate adapter plate 201. This structure utilizes the conductive vias on the adapter plate to achieve vertical interconnection, effectively solving the problem of possible spacing or packaging mismatch between different chips while ensuring electrical connection performance, and improving design flexibility and integration density.

[0075] In optional embodiments of this application, the first receiving processing chip 123 is electrically connected to the switching chip 3, including at least one of the following: 1. For example Figure 2 As shown, the first receiving and processing chip 123 is directly bonded to the switching chip 3, achieving electrical connection. It can be understood that this embodiment eliminates intermediate connection structures by directly bonding the first receiving and processing chip 123 to the switching chip, greatly shortening the electrical signal transmission path. This effectively reduces signal attenuation and parasitic effects, improves data transmission rate and integrity, simplifies packaging processes, and enhances system integration and reliability.

[0076] 2. For example Figure 3 As shown, a second intermediate adapter plate 202 is provided between the first receiving and processing chip 123 and the switching chip 3. The first receiving and processing chip 123 is bonded to the surface of the second intermediate adapter plate 202 facing the first receiving end support plate 121, and the surface of the second intermediate adapter plate 202 facing away from the first receiving end support plate 121 is bonded to the switching chip. Conductive vias are provided on the second intermediate adapter plate 202 to connect corresponding solder points on the two surfaces, thereby realizing the electrical connection between the first receiving and processing chip 123 and the switching chip. It can be understood that this embodiment realizes the heterogeneous integration of the first receiving and processing chip 123 and the switching chip through the second intermediate adapter plate 202. Vertical interconnection is achieved by using the conductive vias on the adapter plate, which effectively solves the problem of chip spacing or package mismatch while ensuring the quality of high-speed signal transmission, and improves design flexibility and integration density.

[0077] In optional embodiments of this application, the second data transfer module 20 is a transmitting data transfer module. For example... Figure 2 As shown, the second data transfer module 20 includes a second light emitting unit 21, a second light receiving unit 22, and a second non-fiber optic light guiding component 23, all disposed on the same side of the carrier (as shown in the circuit board 500). Figure 2As shown, the second light emitting unit 21 includes a second transmitting end support plate 211 and a second laser chip 212 and a second driving chip 213 disposed on the same side of the second transmitting end support plate 211; the second driving chip 213 is electrically connected to the switching chip, and a microcircuit is disposed on the surface of the second transmitting end support plate 211, which is configured to electrically connect the second laser chip 212 and the second driving chip 213; the light emitting surface of the second laser chip 212 faces the circuit board 500 and is configured to emit a target modulated laser beam toward the circuit board 500.

[0078] It is understood that this embodiment discloses the specific structure of the second data transfer module 20, which is used to realize downlink data transmission from the switching chip to the computing chip. Its beneficial effects are as follows: the second optical transmitting unit 21 is electrically connected to the switching chip through the second transmitting end support plate 211, efficiently converting the data processed by the switching chip into optical signals; the second optical receiving unit 22 is electrically connected to the computing chip through the second receiving end support plate 221, accurately restoring the optical signals into electrical data. The light-emitting surface or photosensitive surface of each chip faces the carrier, and together with the second-generation optical fiber guide component 23 disposed on the carrier, the beam is redirected and guided, forming a compact and symmetrical optical interconnect link. This design not only ensures high-speed, low-latency data transmission from the switching chip to the computing chip, but also, together with the first data transfer module 10, constructs a complete bidirectional inter-chip optical communication system, effectively improving the collaborative computing efficiency of the chip cluster.

[0079] like Figure 2 As shown, the second optical receiving unit 22 includes a second receiving end support plate 221 and a second receiving processing chip 223 and a second photoelectric detection chip 222 disposed on the same side of the second receiving end support plate 221; the second photoelectric detection chip 222 is configured to receive a target modulated laser beam, and a microcircuit is disposed on the surface of the second receiving end support plate 221, which is configured to electrically connect the second photoelectric detection chip 222 and the second receiving processing chip 223; the second receiving processing chip 223 is electrically connected to the second computing chip 2.

[0080] In optional embodiments of this application, the second driver chip 213 is electrically connected to the switching chip 3, including at least one of the following: 1. For example Figure 2 As shown, the second driver chip 213 is directly bonded to the switching chip 3 to achieve electrical connection. It can be understood that in this embodiment, the direct bonding between the second driver chip 213 and the switching chip eliminates the intermediate connection structure, greatly shortening the electrical signal transmission path. This effectively reduces signal attenuation and parasitic parameters, improves transmission rate and integrity, simplifies the packaging process, and enhances system integration and reliability.

[0081] 2. For example Figure 3As shown, a second intermediate adapter plate 202 is provided between the second driver chip 213 and the switching chip 3. The second driver chip 213 is bonded to the surface of the second intermediate adapter plate 202 facing the transmitting support plate, and the surface of the second intermediate adapter plate 202 facing away from the transmitting support plate is bonded to the switching chip. Conductive vias are provided on the second intermediate adapter plate 202 to connect corresponding solder points on the two surfaces, thereby realizing the electrical connection between the second driver chip 213 and the switching chip. It can be understood that this embodiment realizes the heterogeneous integration bonding of the second driver chip 213 and the switching chip through the second intermediate adapter plate 202. Vertical interconnection is achieved by using the conductive vias on the adapter plate, which effectively solves the problem of chip spacing or package mismatch while ensuring the quality of high-speed signal transmission, and improves design flexibility and integration density.

[0082] In optional embodiments of this application, the second receiving and processing chip 223 is electrically connected to the second computing chip 2, including at least one of the following: 1. For example Figure 2 As shown, the second receiving and processing chip 223 is directly bonded to the second computing chip 2, achieving electrical connection. It can be understood that this embodiment eliminates intermediate connection structures by directly bonding the second receiving and processing chip 223 to the computing chip, greatly shortening the electrical signal transmission path. This effectively reduces signal attenuation and parasitic effects, improves data transmission rate and integrity, simplifies packaging processes, and enhances system integration and reliability.

[0083] 2. For example Figure 3 As shown, a third intermediate adapter plate 203 is provided between the second receiving and processing chip 223 and the second computing chip 2. The second receiving and processing chip 223 is bonded to the surface of the third intermediate adapter plate 203 facing the second receiving end support plate 221, and the surface of the third intermediate adapter plate 203 facing away from the second receiving end support plate 221 is bonded to the computing chip. Conductive vias are provided on the third intermediate adapter plate 203 to connect corresponding solder points on the two surfaces, thereby realizing the electrical connection between the second receiving and processing chip 223 and the computing chip. It can be understood that this embodiment realizes the heterogeneous integration of the second receiving and processing chip 223 and the computing chip through the third intermediate adapter plate 203. Vertical interconnection is achieved by using the conductive vias on the adapter plate, which effectively solves the problem of chip spacing or package mismatch while ensuring the quality of high-speed signal transmission, and improves design flexibility and integration density.

[0084] In an optional embodiment of this application, the second receiving and processing chip 223 includes a second transimpedance amplifier and a second clock data recovery circuit; the second transimpedance amplifier is configured to receive the second target electrical signal and convert it into a second voltage signal and amplify it; the second clock data recovery circuit is configured to extract the clock and shape the data of the amplified second voltage signal to restore the second target data.

[0085] In an optional embodiment of this application, the non-fiber light guiding component includes a groove disposed on a carrier or other layer structure on the carrier, the groove including at least two reflective surfaces, the at least two reflective surfaces guiding the target modulated laser beam output by the light emitting unit into the light receiving unit.

[0086] In an optional embodiment of this application, the groove includes a first inner sidewall and a second inner sidewall disposed opposite to each other. A first reflective surface is disposed on the first inner sidewall, and a second reflective surface is disposed on the second inner sidewall. The first reflective surface is configured to reflect the target modulated laser beam output by the light emitting unit, causing it to exit in a direction parallel to the carrier surface. The second reflective surface is configured to reflect the target modulated laser beam incident in a direction parallel to the carrier surface, causing it to be guided into the light receiving unit.

[0087] It is understandable that by directly etching a groove with a reflective surface onto the carrier as a non-fiber light guide component, monolithic integration of the optical path structure and the carrier is achieved. This design eliminates the need for assembling independent light guide elements, greatly simplifies the packaging process, improves alignment accuracy and structural stability, while shortening the optical path transmission distance, effectively reducing signal loss and delay, and lowering costs.

[0088] like Figure 2 As shown, the first non-fiber optical guide component in the first data transfer module 10 includes a first groove 135 disposed on the circuit board 500. The first groove 135 includes a first inner sidewall and a second inner sidewall disposed opposite to each other. A first reflective surface 131 is disposed on the first inner sidewall, and a second reflective surface 132 is disposed on the second inner sidewall. The first reflective surface 131 is configured to reflect the target modulated laser beam output by the first light emitting unit 11, so that it is emitted in a direction parallel to the carrier surface (the surface of the circuit board 500). The second reflective surface 121 is configured to reflect the target modulated laser beam incident in a direction parallel to the carrier surface, so that it is guided into the first light receiving unit 12.

[0089] The second non-fiber optical guide component in the second data transfer module 20 includes a second groove 140 disposed on the circuit board 500. The second groove 140 includes a first inner sidewall and a second inner sidewall disposed opposite to each other. A third reflective surface 136 is disposed on the first inner sidewall, and a fourth reflective surface 137 is disposed on the second inner sidewall. The third reflective surface 136 is configured to reflect the target modulated laser beam output by the second light emitting unit 21, so that it is emitted in a direction parallel to the carrier surface (the surface of the circuit board 500). The fourth reflective surface 137 is configured to reflect the target modulated laser beam incident in a direction parallel to the carrier surface, so that it is guided into the second light receiving unit 22.

[0090] As can be understood, this embodiment achieves monolithic integration of the optical path structure and the carrier by directly etching a groove with a reflective surface onto the carrier or its layered structure as a non-fiber light guide component. This design eliminates the assembly steps of independent light guide elements, greatly simplifies the packaging process, improves alignment accuracy and structural stability, and reduces costs.

[0091] In addition, the data transfer module uses a non-fiber optical fiber light guide component for beam guidance, instead of traditional optical fiber. This design has significant advantages: it avoids the alignment difficulties and complex processes of coupling optical fiber to the chip, reducing packaging difficulty and cost; at the same time, the non-fiber optical fiber light guide component can be directly integrated onto the carrier, supporting a more compact chip layout, shortening the optical path transmission distance, and further reducing signal loss and latency.

[0092] Furthermore, the unique design of the optical emitting unit emitting a target-modulated laser beam towards the carrier ensures that the laser beam first strikes the carrier, then is reflected by the non-fiber optic light guide component before being guided into the optical receiving unit. This layout makes full use of the carrier's planar space, avoiding potential heat dissipation interference or space occupation issues that might arise from direct upward laser emission, thus improving the system's integration and thermal management efficiency.

[0093] Secondly, this application discloses an integrated chip, including a carrier and at least one data transfer module as described in any of the first aspects, at least one switching chip, and at least one other chip that generates and / or stores data, disposed on the same side of the carrier.

[0094] Taking computing chips as an example, such as Figure 1 and Figure 2 As shown, a switching chip 3 is disposed between the first computing chip 1 and the second computing chip 2. The first computing chip 1, the second computing chip 2, and the switching chip 3 are all disposed on the same side of the carrier (circuit board 500 in the figure). The first target data generated by the first computing chip 1 is transmitted to the second computing chip through the switching chip 3; the second target data generated by the second computing chip 2 is transmitted to the first computing chip 1 through the switching chip 3.

[0095] It is understandable that integrating multiple computing chips, switching chips, and data transfer modules onto the same carrier and side creates a high-density computing unit. By replacing traditional electrical interconnects with optical interconnects, bandwidth, power consumption, and latency bottlenecks are overcome, effectively aggregating multi-chip computing power, surpassing the physical limits of a single chip, significantly improving inter-chip data exchange efficiency, and providing core support for building high-performance computing clusters.

[0096] In an optional embodiment of this application, the other chips include a memory chip; the switching chip is electrically connected to the memory chip through the data transfer module; when the optical receiving unit of the data transfer module is electrically connected to the switching chip and the optical transmitting unit is electrically connected to the memory chip, the switching chip reads target data from the memory chip; when the optical transmitting unit of the data transfer module is electrically connected to the switching chip and the optical receiving unit is electrically connected to the memory chip, the switching chip stores the target data into the memory chip.

[0097] It is understandable that optical interconnect modules enable bidirectional high-speed data access between switching chips and storage chips. The switching chip can both read and write data to the storage chip, overcoming the bottlenecks of traditional electrical interconnects in terms of storage bandwidth, access latency, and power consumption. This significantly improves data throughput efficiency, achieves efficient collaboration between computing and storage resources, and provides support for the construction of high-density, low-latency compute-memory integrated units.

[0098] In optional embodiments of this application, such as Figure 1 As shown, the integrated chip includes a first receiving data transfer module 10 and a first transmitting data transfer module 20; the switching chip 3 is configured to be electrically connected to the first receiving data transfer module 10 and the first transmitting data transfer module 20 respectively; the end of the first receiving data transfer module 10 away from the switching chip 3 is configured to be electrically connected to the first computing chip 1, and the end of the first transmitting data transfer module 20 away from the switching chip 3 is configured to be electrically connected to the second computing chip 2.

[0099] It is understandable that the first receiver and the first transmitter optical interconnect modules enable efficient unidirectional communication between the switching chip and two different computing chips. The switching chip can acquire data from the first computing chip via the receiver module and simultaneously send data to the second computing chip via the transmitter module, forming a clear pipeline path. This design simplifies optical path layout and signal scheduling, and improves the efficiency of multi-chip collaborative processing.

[0100] In optional embodiments of this application, such as Figure 6 As shown, the integrated chip also includes a second receiving data transfer module 30 and a second transmitting data transfer module 40 electrically connected to the switching chip; the end of the second receiving data transfer module 30 away from the switching chip 3 is configured to be electrically connected to the second computing chip 2, and the end of the second transmitting data transfer module 40 away from the switching chip 3 is configured to be electrically connected to the first computing chip 1.

[0101] It is understandable that by adding a second receiver and a second transmitter module to the first receiver and first transmitter modules, full-duplex bidirectional communication between the switching chip and the first and second computing chips is achieved. The switching chip can simultaneously receive data from and send data to both chips, forming a symmetrical interconnect structure. This design significantly improves data exchange bandwidth and parallel processing capabilities, enhancing the system's communication flexibility.

[0102] In optional embodiments of this application, such as Figure 1 As shown, the switching chip includes a first side L1 and a second side L2 that are arranged opposite to each other. The first computing chip 1 is arranged closer to the first side L1, and the second computing chip 2 is arranged closer to the second side L2.

[0103] It is understandable that placing the first computing chip near the first side of the switching chip and the second computing chip near the opposite second side creates a symmetrical and separated physical layout. This arrangement effectively shortens the transmission distance of the optical interconnect module and reduces signal delay and loss. At the same time, the opposite side layout avoids signal cross-interference, simplifies optical path design and wiring complexity, and helps improve the communication efficiency and integration reliability of multi-chip systems.

[0104] In optional embodiments of this application, such as Figure 7 As shown, the switching chip 3 includes a first side L1 and a third side L3 arranged adjacent to each other. The first computing chip 1 is arranged close to the first side L1, and the second computing chip 2 is arranged close to the third side L3.

[0105] It is understandable that placing the first computing chip near the first side of the switching chip and the second computing chip near the adjacent third side creates a compact L-shaped layout. This arrangement allows for the integration of more chips within a limited space, increasing packaging density. Simultaneously, the adjacent side layout shortens the optical path corner distance, reduces reflection loss and signal delay, and optimizes space utilization and communication efficiency.

[0106] In optional embodiments of this application, the integrated chip includes at least one of the following: The number of first receiving data transfer modules is greater than or equal to 2; The number of the first transmitting data transfer modules is greater than or equal to 2; The number of second receiving data transfer modules is greater than or equal to 2; The number of second-transmit data transfer modules is greater than or equal to 2.

[0107] It is understandable that by configuring multiple first or second receiver-transmitter optical interconnect modules, parallel extended communication between the switching chip and multiple computing chips is achieved. The multi-channel design significantly improves the total bandwidth and data throughput, supporting more complex multi-chip collaborative computing scenarios. At the same time, the flexible configuration of the number of modules enhances the system's scalability, which can be dynamically adjusted according to computing power requirements to meet the large-scale deployment requirements of high-performance clusters.

[0108] like Figure 8 As shown, in addition to the first computing chip 1 and the second computing chip 2, more other chips can be arranged around the switching chip 3, such as the third computing chip 3 and the fourth computing chip 4. In addition, storage chips and other chips can also be arranged.

[0109] In optional embodiments of this application, such as Figure 2 As shown, the switching chip 3, the first computing chip 1, and the second computing chip 2 are all bare chips, and the carrier is the circuit board 500. The bare chips are electrically connected to the surface of the circuit board 500. Non-fiber optic light guide components are disposed in the first groove 135 and the second groove 140 on the circuit board 500.

[0110] As can be understood, this embodiment uses a bare die directly electrically connected to the surface of the circuit board 500, eliminating the need for a traditional packaging shell and greatly shortening the interconnection path. This design effectively reduces signal transmission loss and parasitic effects, improves the integrity of high-speed signals, and simplifies process steps, which is conducive to achieving high-density, low-cost chip integration.

[0111] In optional embodiments of this application, such as Figure 4 As shown, the switching chip 3, the first computing chip 1, and the second computing chip 2 are all bare chips. The carrier is a carrier substrate 200, on which a first circuit board 601, a first light guide carrier plate 602, a second circuit board 603, a second light guide carrier plate 604, and a third circuit board 605 are sequentially disposed. Each bare chip is electrically connected to a single circuit board 500, and each non-fiber optic light guide component is fixed to a single light guide carrier plate. The non-fiber optic light guide component is a groove disposed on the light guide carrier plate, such as the third groove 606 disposed between the first computing chip 1 and the switching chip 3, and the fourth groove 607 disposed between the switching chip 3 and the second computing chip 2.

[0112] The selection of materials for the aforementioned carrier substrate must consider mechanical support, thermal management, and process compatibility. The main materials that may be used include: Silicon substrate: As a preferred option, silicon material has an excellent match with the thermal expansion coefficient of bare die (mostly bare silicon die), which can significantly reduce thermal stress. At the same time, it has high flatness and mature microfabrication technology, which facilitates high-density interconnection.

[0113] Ceramic substrates, such as aluminum nitride (AlN) or aluminum oxide (Al2O3), have good thermal conductivity and insulation properties, making them suitable for high-power applications.

[0114] Organic composite materials, such as epoxy resin or polyimide reinforced substrates, are low-cost and flexible in processing, making them suitable for consumer applications with low heat dissipation requirements.

[0115] Glass substrate: It has excellent high-frequency electrical properties and surface flatness, which is beneficial to signal integrity.

[0116] In summary, silicon substrates, due to their homogeneity with chips, are an ideal basis for high-performance integration.

[0117] It is understood that this embodiment integrates the light guide carrier plate and the circuit board 500 through the carrier substrate, realizing the modular and partitioned assembly of the optical path and the circuit. The bare core is independently electrically connected to the circuit board 500, and the non-fiber light guide components are independently fixed to the light guide carrier plate, which reduces the difficulty of alignment and coupling, improves the assembly yield and maintainability, and facilitates the separate optimization of the optical path and the circuit.

[0118] In optional embodiments of this application, the other chip is a calculator circuit board carrying a bare computing chip and its peripheral auxiliary circuits; the switching chip is a switch circuit board carrying a bare switching chip and its peripheral auxiliary circuits; the carrier is a substrate, and the calculator circuit board and the switch circuit board are fixed on the substrate; the non-fiber optic light guide component is a groove disposed on the substrate.

[0119] In optional embodiments of this application, the other chip is a memory circuit board carrying a bare memory chip and its peripheral auxiliary circuits; the switching chip is a switch circuit board carrying a bare switching chip and its peripheral auxiliary circuits; the carrier is a substrate, and the memory circuit board and the switch circuit board are fixed on the substrate; the non-fiber optic light guide component is a groove disposed on the substrate.

[0120] In optional embodiments of this application, the peripheral auxiliary circuit includes at least one of the following: a power management circuit, a clock circuit, a data interface circuit, and a storage circuit.

[0121] It is understandable that the computing or storage bare chip and its peripheral auxiliary circuitry are integrated onto a separate circuit board, then fixed to a substrate, and optical interconnects are achieved using substrate recesses. This modular design facilitates independent testing, replacement, and maintenance of each component, reducing alignment difficulties and packaging costs. Simultaneously, the proximity of peripheral circuitry optimizes power integrity and signal quality, improves system reliability and integration flexibility, and is beneficial for mass production.

[0122] like Figure 5As shown, there is a first calculator circuit board 701 containing a first computing chip die and peripheral auxiliary circuits, a second calculator circuit board 702 containing a second computing chip die and peripheral auxiliary circuits, and a switch circuit board 703 containing a switch chip die and peripheral auxiliary circuits.

[0123] like Figure 5 As shown, the carrier is a silicon substrate 700, and the first calculator circuit board 701, the second calculator circuit board 702 and the switch circuit board 703 are fixed on the substrate 700.

[0124] It is understood that this embodiment achieves modular integration by pre-packaging the computing chip and the switching chip into independent circuit board modules, and then fixing them uniformly on the substrate carrier. Each module can be tested and screened independently, improving yield and maintainability; the integration of peripheral auxiliary circuits (such as power supply and clock) optimizes the chip's operating environment; at the same time, the modular design facilitates flexible expansion and heterogeneous integration, reducing the complexity of system-level packaging.

[0125] In optional embodiments of this application, the peripheral auxiliary circuitry includes at least one of the following: a power management circuit, a clock circuit, a data interface circuit, and a storage circuit.

[0126] The material selection for the aforementioned substrate focuses on structural strength, heat dissipation, and signal transmission. The main materials that may be used include: Silicon substrate: It has excellent thermal stability and a thermal expansion coefficient that matches the chip, which can accurately support the circuit board and dissipate heat efficiently, making it suitable for high-performance computing scenarios.

[0127] Metal-based composite materials, such as aluminum-based or copper-based substrates, have excellent thermal conductivity and can quickly dissipate the heat generated by the calculator circuit board, making them suitable for high-power applications.

[0128] Ceramic substrates, such as aluminum nitride, provide good insulation and thermal conductivity, ensuring stable high-frequency signal transmission.

[0129] Organic laminates: such as FR-4 epoxy glass cloth laminate, are low in cost and have mature processing, and are widely used in general electronic equipment.

[0130] In summary, silicon substrates provide a solid platform foundation for high-performance multi-chip systems due to their high reliability and integration potential.

[0131] In an optional embodiment of this application, a non-fiber optic light guide component is disposed between two adjacent calculator circuit boards on the substrate, such as... Figure 5 As shown, a fifth groove 704 is provided between the first calculator circuit board 701 and the second calculator circuit board 702, and a sixth groove 705 is provided between the second calculator circuit board 702 and the third calculator circuit board 703.

[0132] Thirdly, this application discloses a vertically scalable network architecture, which employs a computing power cluster composed of multiple integrated chips as described in any one of the second aspects.

[0133] The vertically scalable network architecture disclosed in the third aspect of this application can be a scale-up architecture. A scale-up architecture refers to improving system performance by enhancing the computing power of individual nodes (such as increasing CPU / memory or integrating more chips), rather than increasing the number of nodes. The scale-up architecture in this application uses inter-chip optical interconnect technology to interconnect multiple integrated chips at high speed, aggregating computing power and enabling them to work collaboratively like a powerful "single computer" to meet the demands of high-performance computing. This architecture, by employing multiple of the aforementioned integrated chips to build a high-density computing cluster, aims to meet the core requirements of computing power aggregation and efficient collaboration in large-scale parallel computing scenarios.

[0134] In a scale-up architecture, multiple integrated chips are interconnected via chip-level optical communication through a data transfer module, forming a tightly coupled pool of computing resources. Optical interconnects offer advantages such as high bandwidth, low latency, and interference resistance, effectively overcoming the signal attenuation and power consumption bottlenecks of traditional electrical interconnects during expansion, and supporting high-speed, stable data exchange between chips. This architecture supports flexible scheduling of various computing resources and is widely applicable to application scenarios with extremely high requirements for computing power aggregation, such as artificial intelligence training, high-performance computing, and real-time big data analysis.

[0135] Furthermore, this vertically scaling architecture boasts excellent modularity and scalability, enabling dynamic adjustment of the cluster size based on computing power demands, thereby achieving elastic deployment and efficient utilization of computing resources. This design provides an engineering-feasible technical path for building next-generation high-performance computing platforms.

[0136] The terms "first," "second," "first," or "second" as used in the various embodiments of this disclosure may modify various components regardless of their order and / or importance, but these terms do not limit the corresponding components. The above terms are configured only for the purpose of distinguishing an element from other elements. For example, "first user equipment" and "second user equipment" refer to different user equipments, although both are user equipment. For example, without departing from the scope of this disclosure, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element.

[0137] When a component (e.g., a first component) is referred to as being "(operably or communicatively) coupled" or "(operably or communicatively) coupled to" or "connected to" another component (e.g., a second component), it should be understood that the first component is directly connected to the second component or that the first component is indirectly connected to the second component via yet another component (e.g., a third component). Conversely, it can be understood that when a component (e.g., a first component) is referred to as being "directly connected" or "directly coupled" to another component (the second component), no component (e.g., a third component) is inserted between the two.

[0138] It should be noted that, in this document, 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 that element. Furthermore, components, features, and elements with the same names in different embodiments of this application may have the same meaning or different meanings, the specific meaning of which must be determined by its interpretation in that specific embodiment or further in conjunction with the context of that specific embodiment.

[0139] The above description is merely an optional embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

[0140] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”

[0141] The above description is merely an optional embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

[0142] The above description is merely an optional embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A data flow conversion module, characterized by: Includes a light emitting unit, a light receiving unit, and a non-fiber optic light guide assembly, all located on the same side of the carrier; When the optical receiving unit is electrically connected to the switching chip, the optical transmitting unit is electrically connected to other chips that generate and / or store data; or, when the optical transmitting unit is electrically connected to the switching chip, the optical receiving unit is electrically connected to other chips that generate and / or store data. The non-fiber optical guide component is disposed on the carrier surface and located between the other chips and the switching chip, and is configured to guide the target modulated laser beam output by the optical emitting unit into the optical receiving unit through at least two levels of specular reflection in free space; The data transfer module includes a receiving data transfer module and a transmitting data transfer module; The optical receiving unit in the receiving data transfer module is electrically connected to the switching chip, and the optical transmitting unit in the receiving data transfer module is electrically connected to other chips that generate and / or store data. The optical transmitting unit in the transmitting data transfer module is electrically connected to the switching chip, and the optical receiving unit in the transmitting data transfer module is electrically connected to other chips that generate and / or store data. The light emitting unit includes a transmitting end support plate and a laser chip and a driving chip disposed on the same side of the transmitting end support plate; the driving chip is electrically connected to the switching chip or the other chip; a microcircuit is disposed on the surface of the transmitting end support plate, and the microcircuit is configured to electrically connect the laser chip and the driving chip; the light emitting surface of the laser chip faces the carrier and is configured to emit a target modulated laser beam toward the carrier; The optical receiving unit includes a receiving end support plate and a receiving processing chip and a photodetector chip disposed on the same side of the receiving end support plate; the photodetector chip is configured to receive the target modulated laser beam, and a microcircuit is disposed on the surface of the receiving end support plate, the microcircuit being configured to electrically connect the photodetector chip and the receiving processing chip; the receiving processing chip is electrically connected to the other chips or the switching chip.

2. The data transfer module according to claim 1, characterized in that, The non-fiber optical guide component includes a groove disposed on a carrier or other layer structure on the carrier, the groove including at least two reflective surfaces, the at least two reflective surfaces guiding the target modulated laser beam output by the optical emitting unit into the optical receiving unit.

3. The data transfer module according to claim 2, characterized in that, The groove includes a first inner sidewall and a second inner sidewall disposed opposite to each other, wherein a first reflective surface is disposed on the first inner sidewall and a second reflective surface is disposed on the second inner sidewall; The first reflective surface is configured to reflect the target modulated laser beam output by the light emitting unit, causing it to exit in a direction parallel to the carrier surface; The second reflective surface is configured to reflect the target modulated laser beam incident in a direction parallel to the carrier surface, and guide it into the light receiving unit.

4. The data transfer module according to claim 1, characterized in that, The driver chip is electrically connected to the switching chip or the other chip, including at least one of the following: The driver chip is directly bonded to the switching chip or other chips to achieve electrical connection; An intermediate adapter board is provided between the driver chip and the switching chip or the other chip. The driver chip is bonded to the surface of the intermediate adapter board facing the transmitter support plate, and the surface of the intermediate adapter board facing away from the transmitter support plate is bonded to the switching chip or the other chip. The intermediate adapter board is provided with conductive vias to connect corresponding solder joints on the two surfaces, so as to realize the electrical connection between the driver chip and the switching chip or the other chip.

5. The data transfer module according to claim 1, characterized in that, The receiving processing chip is electrically connected to the other chips or the switching chip, including at least one of the following: The receiving and processing chip is directly bonded to the other chips or the switching chip to achieve electrical connection; An intermediate adapter board is provided between the receiving and processing chip and the other chips or the switching chip. The receiving and processing chip is bonded to the surface of the intermediate adapter board facing the receiving end support plate, and the surface of the intermediate adapter board facing away from the receiving end support plate is bonded to the other chips or the switching chip. The intermediate adapter board is provided with conductive vias to connect corresponding solder joints on the two surfaces, so as to realize the electrical connection between the receiving and processing chip and the other chips or the switching chip.

6. The data transfer module according to any one of claims 1 to 5, characterized in that, The other chips include at least one of the following: a computing chip or a storage chip.

7. An integrated chip, characterized by It includes a carrier and at least one data transfer module as described in any one of claims 1-6, at least one switching chip, and at least one other chip that generates and / or stores data, disposed on the same side of the carrier.

8. The integrated chip according to claim 7, characterized in that, The other chips include memory chips; The switching chip is electrically connected to the storage chip through the data transfer module; When the optical receiving unit of the data transfer module is electrically connected to the switching chip and the optical transmitting unit is electrically connected to the storage chip, the switching chip reads the target data from the storage chip. When the optical transmitting unit of the data transfer module is electrically connected to the switching chip and the optical receiving unit is electrically connected to the storage chip, the switching chip stores the target data into the storage chip.

9. The integrated chip of claim 7, wherein, It includes a first receiving data transfer module and a first transmitting data transfer module; The switching chip is configured to be electrically connected to the first receiving data transfer module and the first transmitting data transfer module, respectively. The end of the first receiving data transfer module opposite to the switching chip is configured to be electrically connected to the first computing chip, and the end of the first transmitting data transfer module opposite to the switching chip is configured to be electrically connected to the second computing chip.

10. The integrated chip according to claim 9, characterized in that, The number of the first receiving data transfer modules is greater than or equal to 2; The number of the first transmitting data transfer modules is greater than or equal to 2.

11. The integrated chip according to claim 7, characterized in that, It also includes a second receiving data transfer module and a second transmitting data transfer module that are electrically connected to the switching chip; The end of the second receiving data transfer module opposite to the switching chip is configured to be electrically connected to the second computing chip, and the end of the second transmitting data transfer module opposite to the switching chip is configured to be electrically connected to the first computing chip.

12. The integrated chip according to claim 11, characterized in that, The number of the second receiving data transfer modules is greater than or equal to 2; The number of the second transmitting data transfer modules is greater than or equal to 2.

13. The integrated chip according to claim 9 or 11, characterized in that, The switching chip includes a first side and a second side disposed opposite to each other, with the first computing chip disposed closer to the first side and the second computing chip disposed closer to the second side.

14. The integrated chip according to claim 9 or 11, characterized in that, The switching chip includes a first side and a third side arranged adjacent to each other, with the first computing chip arranged close to the first side and the second computing chip arranged close to the third side.

15. The integrated chip according to any one of claims 7-11, characterized in that, The switching chip and the other chips are all bare cores, the carrier is a circuit board, the bare cores are electrically connected to the surface of the circuit board, and the non-fiber optical guide components are disposed in the grooves on the circuit board.

16. The integrated chip according to any one of claims 7-11, characterized in that, The switching chip and the other chips are all bare chips. The carrier is a carrier substrate, and at least one light guide carrier plate and at least two circuit boards are disposed on the carrier substrate. Each of the bare cores is electrically connected to a single circuit board. Each of the non-fiber optic light guide components is fixed to a single light guide carrier plate, wherein the non-fiber optic light guide component is a groove disposed on the light guide carrier plate.

17. The integrated chip according to any one of claims 7-11, characterized in that, The other chips are calculator circuit boards carrying a bare computing chip and its peripheral auxiliary circuits; the switching chip is a switch circuit board carrying a bare switching chip and its peripheral auxiliary circuits; the carrier is a substrate, and the calculator circuit board and the switch circuit board are fixed on the substrate; the non-fiber optic light guide component is a groove disposed on the substrate.

18. The integrated chip according to any one of claims 7-11, characterized in that, The other chips are memory circuit boards carrying bare memory cores and their peripheral auxiliary circuits; the switching chip is a switch circuit board carrying bare switching cores and their peripheral auxiliary circuits; the carrier is a substrate, and the memory circuit board and the switch circuit board are fixed on the substrate; the non-fiber light guide component is a groove disposed on the substrate.

19. A scale-out network architecture, comprising: The architecture employs multiple integrated chips as described in any one of claims 7-18 to form a computing power cluster.

Citation Information

Patent Citations

  • Micro-LED-based optical interconnection chip and packaging method

    CN120891595A

  • Optical Chip-to-Chip Interconnect and Method of Integration

    US20250216635A1