Communication method and device for interconnecting and packaging memristor core particles and silicon-based core particles

By adopting handshake protocol and hierarchical timing control in the Chiplet system, the problem of low communication and synchronization speed between different core particles is solved, and efficient task coordination and system performance improvement is achieved.

CN118467449BActive Publication Date: 2024-11-08TSINGHUA UNIVERSITY
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202410532268.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-29
Publication Date
2024-11-08
Estimated Expiration
2044-04-29

AI Technical Summary

Technical Problem

In the current Chiplet system, the communication and data synchronization speeds between different core particles are relatively low, making it difficult to balance performance, power consumption and cost, and the combination of core particles with different processes and control timings is difficult to effectively coordinate.

Method used

A communication method for interconnection packaging of memristor core particles and silicon-based core particles is proposed. Through the handshake protocol, the timing control of cycle and phase levels is realized, the task coordination and data synchronization are ensured, and the task coordination efficiency between different core particles is improved. The advantages of fusing two core particles through a packaging scheme.

Benefits of technology

It realizes efficient communication and synchronization between different core particles, improves system performance, optimizes resource utilization and timing characteristics of circuit design, and enhances the synchronization mechanism of multi-core/multi-core processors.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118467449B_ABST
    Figure CN118467449B_ABST
Patent Text Reader

Abstract

The present application proposes a communication method and device for interconnecting and packaging memristive core particles and silicon-based core particles, wherein the method includes: hierarchically dividing the working timing of memristive units and silicon-based units, respectively implementing a cycle (step) timing control unit and a phase (phase) timing control unit inside the memristive chip and the silicon-based chip, dividing the task timing for any size task, dividing the task corresponding to the cycle (step) into high-level and coarse granularity, and dividing the corresponding phase (phase) into low-level timing and fine granularity, setting the handshake protocol of the memristive unit and the silicon-based unit, exchanging data and signaling according to the handshake protocol, realizing communication, and completing the coordination and synchronization requirements of interactive tasks. The present invention using the above scheme can give full play to the advantages of memristive core particles and silicon-based core particles, realize the synchronization mechanism of two different types of core particles, improve the task coordination efficiency between different core particles, and realize the improvement of overall system performance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of communication system processing technology, and in particular to a communication method and device for interconnecting and packaging memristive core particles and silicon-based core particles. Background Art

[0002] In the future, integrated circuits will develop towards multi-die integration, that is, multiple verified, unpackaged chip components with different functions will be interconnected and assembled, and packaged as a whole chip in the same tube shell to form a multi-chip module (MCM). The bare cores that make up the MCM are called chiplets.

[0003] Neuromorphic computing core particles based on large-scale memristor arrays can realize high-speed matrix operations in the analog domain. The memristor particles themselves have the characteristics of integrated storage and computing, which makes a single memristor particle have pipeline characteristics, while silicon-based smart core particles can provide more flexible general computing in the digital domain. The storage and computing of silicon-based particles are separated, and the storage has address characteristics.

[0004] Compared with the high-speed and timely communication and data transmission between modules on traditional integrated circuits, the current chiplet bare chips have a low transmission speed for interconnection and communication on the substrate. At the same time, the chiplet needs to properly combine the design target function with the existing chiplet types to reasonably arrange the placement of the chiplets on the substrate and the data communication method, in order to effectively achieve a balance between performance, power consumption and cost.

[0005] In actual applications, the realization of chiplet system functions depends on the collaborative work and interconnected communication between different chiplets. Therefore, there are needs for communication and data synchronization between different chiplets. Furthermore, achieving high-speed interconnection between chiplets and efficient transmission of information within the chiplet system is a top priority in the development of chiplet technology. Summary of the invention

[0006] The present application aims to solve one of the technical problems in the related art at least to some extent.

[0007] To this end, the first purpose of the present application is to propose a communication method for interconnecting and packaging memristive core particles and silicon-based core particles, which solves the current technical problem that circuits with different processes and different control timings have not been combined together in the form of core particles, and realizes the fusion of memristive core particles and silicon-based core particles. It can give full play to the advantages of memristive core particles and silicon-based core particles, realize the synchronization mechanism of two different types of core particles, improve the task coordination efficiency between different core particles, and realize the improvement of overall system performance.

[0008] The second objective of the present application is to provide a communication device in which a memristor chip and a silicon-based chip are interconnected and packaged.

[0009] To achieve the above-mentioned purpose, the first embodiment of the present application proposes a communication method for interconnecting and packaging a memristive core particle and a silicon-based core particle, wherein the memristive core particle and the silicon-based core particle implement a communication handshake according to a handshake protocol, and the method comprises: hierarchically dividing the working timing of the memristive core particle and the silicon-based core particle according to a cycle (step) level and a phase (phase) level, and implementing cycle (step) timing control and phase (phase) timing control inside the memristive core particle and the silicon-based core particle, wherein a cycle (step) includes multiple phases (phase); obtaining tasks, and dividing them in terms of task timing, dividing the cycle (step) into two phases; The tasks corresponding to each ep are divided into high-level timings, and the corresponding phases are divided into low-level timings, wherein the cycle (step) is the maximum granularity of the handshake between the chiplets, and the phase (phase) is the minimum granularity of the handshake between the chiplets; when the task is executed, at each cycle (step) stage, the memristive chiplet or the silicon-based chiplet acts as the main control unit to start the cycle (step), and at each phase (phase) stage, the memristive chiplet or the silicon-based chiplet completes the divided tasks, and transmits control signals or data between the memristive chiplet and the silicon-based chiplet to complete the coordination and synchronization of the interactive tasks.

[0010] The communication method for interconnecting and packaging memristive core particles and silicon-based core particles in the embodiment of the present application integrates memristive core particles and silicon-based core particles together with a packaging solution, and provides a timing handshake protocol, which can realize the interconnection handshake between memristive core particles and silicon-based core particles, provide the operation and management of the communication between memristive core particles and silicon-based core particles, and improve the implementation efficiency and portability; during the handshake, it can realize the writing of the shortest path handshake protocol based on the minimum transmission delay, thereby achieving better resource utilization and effectively improving the timing characteristics of the circuit design.

[0011] Optionally, in one embodiment of the present application, the interaction requirements include data exchange and control signaling exchange, wherein the data exchange is completed by the memristive core particle and the silicon-based core particle, including:

[0012] In a single phase, the memristor chip processes the data and transmits it to the silicon chip. After the silicon chip receives the complete data and completes the data transmission, it enters the next phase, and the silicon information initiates the next operation. Or,

[0013] In a single phase, the memristor core processes the complete data transmitted by the silicon-based core, and after the data transmission is completed, it enters the next phase, and the memristor core initiates the next operation.

[0014] The control signaling exchange is completed by the memristive chip and the silicon-based chip, including:

[0015] In a single phase, the control signal is transmitted to the silicon-based chip through the memristor chip, and after the transmission is completed, the next phase is entered, and the silicon-based chip enters the next operation according to the control signal content, or,

[0016] In a single phase timing, the control signal transmitted by the silicon-based chip is received by the memristive chip, and after the transmission is completed, it enters the next phase timing, and the next operation is performed according to the signal content through the memristive chip.

[0017] Optionally, in one embodiment of the present application, data synchronization is performed periodically between the memristive core and the silicon-based core through signals within the chip, so that the memristive core and the silicon-based core share data.

[0018] Optionally, in one embodiment of the present application, the control signaling in the handshake protocol includes: cycle (step)_start, cycle (step)_running, cycle (step)_progress, cycle (step)_finish, cycle (step)_ack, and the corresponding phase (phase) level includes phase (phase)_start, phase (phase)_running, phase (phase)_progress, phase (phase)_finish, phase (phase)_ack.

[0019] Optionally, in one embodiment of the present application, the handshake protocol process of the loop (step) includes:

[0020] When the task is executed, at each cycle (step) stage, the memristor core particle or silicon-based core particle as the main control unit sends a cycle (step)_start control signal to start the cycle (step), and the silicon-based core particle or memristor core particle receives the cycle (step)_start control signal and feeds back the cycle (step)_ack control signal;

[0021] When the task is executed, at each cycle (step) stage, the memristive core or silicon-based core sends a cycle (step)_finish control signal to end the cycle (step), and the silicon-based core or memristive core receives the cycle (step)_finish control signal and feeds back the cycle (step)_ack control signal.

[0022] Optionally, in one embodiment of the present application, the handshake protocol process of the phase includes:

[0023] When the task is executed, in each cycle (step) stage, after the cycle (step) is started, each phase (phase) stage is entered in sequence;

[0024] In each phase, the memristor chip or silicon-based chip sends a phase_start control signal to start the phase, and the silicon-based chip or memristor chip receives the phase_start control signal and feeds back the phase_ack control signal.

[0025] In each phase stage, the memristive chip or silicon-based chip sends a phase_finish control signal to end the phase, and the silicon-based chip or memristive chip receives the phase_finish control signal and feeds back a phase_ack control signal.

[0026] To achieve the above-mentioned purpose, the second aspect of the present invention proposes a communication device for interconnecting and packaging a memristive core and a silicon-based core, comprising a memristive core and a silicon-based core, wherein the memristive core and the silicon-based core implement a communication handshake according to a handshake protocol, the working timing of the memristive core and the silicon-based core is hierarchically divided according to a cycle (step) level and a phase (phase) level, and a cycle (step) timing control unit and a phase (phase) timing control unit are implemented inside the memristive core and the silicon-based core, and the cycle (step) includes multiple phases (phase);

[0027] When a task is executed, in each cycle (step) stage, the memristive chip or silicon-based chip acts as the main control unit to start the cycle (step), and in each phase (phase) stage, the memristive chip or silicon-based chip completes the divided tasks and transmits control signals or data to complete the coordinated synchronization of interactive tasks, wherein the task is a task divided in task timing, and when dividing, the task corresponding to the cycle (step) is divided into a high-level timing, and the corresponding phase (phase) is divided into a low-level timing, the cycle (step) is the maximum granularity of the handshake between the chiplets, and the phase (phase) is the minimum granularity of the handshake between the chiplets.

[0028] Optionally, in one embodiment of the present application, the interaction requirements include data exchange and control signaling exchange, wherein the memristive core particle and the silicon-based core particle complete the data exchange, including:

[0029] In a single phase, the memristor chip completes the data processing and transmits the data to the silicon chip. The silicon chip receives the complete data and enters the next phase after the data transmission is completed. The silicon information starts the next operation, or,

[0030] In a single phase, the memristor chip processes the complete data transmitted by the silicon chip, and after the data transmission is completed, it enters the next phase, and the memristor chip starts the next operation.

[0031] The memristive core and the silicon-based core complete the control signaling exchange, including:

[0032] In a single phase, the memristor chip transmits control signals to the silicon chip, and after the transmission is completed, it enters the next phase, and the silicon chip proceeds to the next step according to the control signal content, or,

[0033] In a single phase timing, the memristor chip receives the control signal transmitted by the silicon-based chip, and after the transmission is completed, it enters the next phase timing, and the memristor chip proceeds to the next step according to the signal content.

[0034] Optionally, in one embodiment of the present application, the memristive core and the silicon-based core periodically perform data synchronization via signals within the chip, and the memristive core and the silicon-based core share data.

[0035] Optionally, in one embodiment of the present application, the control signaling in the handshake protocol includes: cycle (step)_start, cycle (step)_running, cycle (step)_progress, cycle (step)_finish, cycle (step)_ack, and the corresponding phase (phase) level includes phase (phase)_start, phase (phase)_running, phase (phase)_progress, phase (phase)_finish, phase (phase)_ack.

[0036] The communication device for interconnecting and packaging memristive core particles and silicon-based core particles in the embodiment of the present application proposes coarse-grained synchronization mechanisms and fine-grained synchronization mechanisms at different levels to achieve task synchronization between memristive core particles and silicon-based core particles, and provides a timing handshake protocol to achieve interconnection handshake between memristive core particles and silicon-based core particles. Through the proposed handshake protocol and synchronization mechanism supported by on-chip dedicated hardware, the synchronization mechanism of multi-core / many-core processing processors is significantly improved.

[0037] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0039] Figure 1 A flow chart of a communication method for interconnecting and packaging a memristor core particle and a silicon-based core particle provided in the first embodiment of the present application;

[0040] Figure 2 A schematic diagram of a heterogeneous core grain architecture design according to an embodiment of the present application;

[0041] Figure 3 A schematic diagram of a handshake processing solution according to an embodiment of the present application;

[0042] Figure 4 A schematic diagram of the hierarchical division of an embodiment of the present application;

[0043] Figure 5 A schematic diagram of a handshake protocol process for a single data exchange and a single signaling exchange according to an embodiment of the present application;

[0044] Figure 6 A schematic diagram of the structure of a communication device in which a memristor core and a silicon-based core are interconnected and packaged, provided in an embodiment of the present application. DETAILED DESCRIPTION

[0045] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0046] Neuromorphic computing systems can integrate time, space, phase, and frequency information through a variety of pulse coding schemes, and events are sparse and have high capacity. Due to the introduction of the time dimension, neuromorphic computing has more powerful computing capabilities, and the sparse nature of events also brings the advantage of low power consumption to the system. The inherent characteristics of the brain create the high energy efficiency and intelligent computing characteristics of neuromorphic computing. In general, neuromorphic computing has the advantages of high parallelism, storage and computing fusion, analog computing, high plasticity, high scalability, and autonomous learning.

[0047] Further combining the high parallelism and low power consumption of the memristor storage and computing integrated chip with the high-speed general-purpose processing capability of the silicon-based intelligent computing chip, and the on-chip communication interaction protocol between the memristor chip and the silicon-based chip, we can ultimately realize a heterogeneous, energy-efficient hybrid digital-analog artificial intelligence chip and build a chip design platform based on chiplet technology.

[0048] Neuromorphic computing chips based on traditional silicon-based transistors and new neuromorphic computing chips based on emerging nano-memristors.

[0049] The advantage of chiplet technology is that it can provide better manufacturing flexibility and scalability. Since the main chip is decomposed into multiple small chiplets, each chiplet can be manufactured and tested independently.

[0050] The synchronization mechanism between multiple cores is the key to the correct execution and coordinated communication of multi-core / many-core processing units, and its efficiency is very important to the performance of the overall system. The proposed and implemented coarse-grained synchronization mechanism and fine-grained synchronization mechanism at different levels, namely the handshake protocol and synchronization mechanism supported by dedicated hardware on the chip, significantly improve the synchronization mechanism of multi-core / many-core processing processors.

[0051] The following describes the communication method and device for interconnecting and packaging a memristive chip and a silicon-based chip in an embodiment of the present application with reference to the accompanying drawings.

[0052] Figure 1 A flow chart of a communication method for interconnecting and packaging a memristive core particle and a silicon-based core particle provided in the first embodiment of the present application, wherein the memristive core particle and the silicon-based core particle implement a communication handshake according to a handshake protocol.

[0053] like Figure 1 As shown, the communication method of the memristor core and the silicon-based core interconnect package includes the following steps:

[0054] Step 101, hierarchically dividing the working timing of the memristor core particles and the silicon-based core particles according to the cycle (step) level and the phase (phase) level, and realizing the cycle (step) timing control and the phase (phase) timing control inside the memristor core particles and the silicon-based core particles, wherein the cycle (step) includes multiple phases (phase);

[0055] Step 102, obtain tasks and divide them in terms of task timing, divide the tasks corresponding to the cycle (step) into high-level timing, and divide the corresponding phase (phase) into low-level timing, where the cycle (step) is the maximum granularity of handshake between core particles, and the phase (phase) is the minimum granularity of handshake between core particles;

[0056] Step 103, when the task is executed, in each cycle (step) stage, the memristive chip or the silicon-based chip acts as the main control unit to start the cycle (step), and in each phase (phase) stage, the memristive chip or the silicon-based chip completes the divided tasks, and transmits control signals or data between the memristive chip and the silicon-based chip to complete the coordination and synchronization of the interactive tasks.

[0057] Specifically, through the communication handshake between the memristor chip and the silicon-based chip, the coordination and synchronization requirements of the pipeline properties or address properties of the silicon-based chip and the pipeline properties of the memristor chip are met.

[0058] The communication method for interconnecting and packaging memristive core particles and silicon-based core particles in the embodiment of the present application integrates memristive core particles and silicon-based core particles together with a packaging solution, and provides a timing handshake protocol, which can realize the interconnection handshake between memristive core particles and silicon-based core particles, provide the operation and management of the communication between memristive core particles and silicon-based core particles, and improve the implementation efficiency and portability; during the handshake, it can realize the writing of the shortest path handshake protocol based on the minimum transmission delay, thereby achieving better resource utilization and effectively improving the timing characteristics of the circuit design.

[0059] Optionally, in an embodiment of the present application, the interaction requirement includes data exchange and control signaling exchange, wherein the data exchange is completed by the memristive core particle and the silicon-based core particle, including:

[0060] In a single phase, the memristor chip processes the data and transmits it to the silicon chip. After the silicon chip receives the complete data and completes the data transmission, it enters the next phase, and the silicon information initiates the next operation. Or,

[0061] In a single phase, the memristor core processes the complete data transmitted by the silicon-based core, and after the data transmission is completed, it enters the next phase, and the memristor core initiates the next operation.

[0062] The control signaling exchange is completed by the memristive chip and the silicon-based chip, including:

[0063] In a single phase, the control signal is transmitted to the silicon-based chip through the memristor chip, and after the transmission is completed, the next phase is entered, and the silicon-based chip enters the next operation according to the control signal content, or,

[0064] In a single phase timing, the control signal transmitted by the silicon-based chip is received by the memristive chip, and after the transmission is completed, it enters the next phase timing, and the next operation is performed according to the signal content through the memristive chip.

[0065] Optionally, in one embodiment of the present application, it further includes:

[0066] The connection between memristor core particles and silicon-based chips is needed to synchronize data between multiple chips. Synchronization can keep the data between two or more units consistent and tasks mutually synchronized, thereby ensuring the accuracy and reliability of the packaging system, and meeting the needs of real-time response systems. It synchronizes the signals within the chip regularly to keep the timing signals between chips consistent. At the same time, the technology between the two core particles can also be used for data sharing between two or more units, thereby realizing resource sharing between multiple units, greatly improving the processing efficiency and parallelism of the system.

[0067] Optionally, in one embodiment of the present application, the control signaling in the handshake protocol includes: cycle (step)_start, cycle (step)_running, cycle (step)_progress, cycle (step)_finish, cycle (step)_ack, and the corresponding phase (phase) level includes phase (phase)_start, phase (phase)_running, phase (phase)_progress, phase (phase)_finish, phase (phase)_ack.

[0068] Optionally, in one embodiment of the present application, the handshake protocol process of the loop (step) includes:

[0069] When the task is executed, at each cycle (step) stage, the memristor core particle or silicon-based core particle as the main control unit sends a cycle (step)_start control signal to start the cycle (step), and the silicon-based core particle or memristor core particle receives the cycle (step)_start control signal and feeds back the cycle (step)_ack control signal;

[0070] When the task is executed, at each cycle (step) stage, the memristive core or silicon-based core sends a cycle (step)_finish control signal to end the cycle (step), and the silicon-based core or memristive core receives the cycle (step)_finish control signal and feeds back the cycle (step)_ack control signal.

[0071] Optionally, in one embodiment of the present application, the handshake protocol process of the phase includes:

[0072] When the task is executed, in each cycle (step) stage, after the cycle (step) is started, each phase (phase) stage is entered in sequence;

[0073] In each phase, the memristor chip or silicon-based chip sends a phase_start control signal to start the phase, and the silicon-based chip or memristor chip receives the phase_start control signal and feeds back the phase_ack control signal.

[0074] In each phase stage, the memristive chip or silicon-based chip sends a phase_finish control signal to end the phase, and the silicon-based chip or memristive chip receives the phase_finish control signal and feeds back a phase_ack control signal.

[0075] Figure 2 Schematic diagram of the heterogeneous core grain architecture design of this embodiment. Figure 2 As shown in the figure, it covers two different intelligent computing cores, one is a neuromorphic computing core based on a large-scale memristor array, and the other is a silicon-based intelligent computing core. Through the mixed digital-analog computing method, the advantages of the two different computing cores are integrated, so as to provide a flexible combination for the needs of different scenarios. The memristor neuromorphic computing core can realize high-speed matrix operations in the analog domain, while the silicon-based intelligent core can provide more flexible general computing in the digital domain. Their collaborative work can make the entire computing system more efficient and reliable. Through reasonable interconnection and communication technology, the advantages of these two cores can be fully utilized to improve the overall system performance.

[0076] Chiplet technology, on-chip interconnect technology and multi-dimensional communication networking integrate two different chiplets, giving full play to the advantages of the two chiplets, thereby improving the overall performance of the system.

[0077] This embodiment provides Figure 3 The handshake processing scheme shown can provide in-depth understanding of the execution characteristics of the algorithm on different computing core particles, including comprehensive analysis of running time, memory access mode, operation combination, data representation, etc. Implement core particle management and programming model. Core particle management and programming models can be selected and customized according to specific chip design and application requirements, simplify core particle management and programming operations, improve implementation efficiency and system performance, and give full play to the flexibility and scalability of chip design platforms based on chiplets. A distributed programming model is adopted, and communication mechanisms such as shared memory and layer 2 handshake are used to realize data exchange and collaborative work between core particles, and the computing and communication tasks of different core particles are decomposed into smaller subtasks to improve parallelism and efficiency.

[0078] like Figure 4As shown, the working timing of the memristor unit and the silicon-based unit is hierarchically divided. According to the phase and cycle levels, a cycle contains a maximum of 64 phases, and the cycle timing control unit and the phase timing control unit are implemented in the memristor chip and the silicon-based chip respectively. The control signaling in the handshake protocol includes cycle_start, cycle_running, cycle_progress, cycle_finish, and cycle_ack, and the corresponding phase level includes phase_start, phase_running, phase_progress, phase_finish, and phase_ack. Based on the above signaling, two communication handshakes with different processes and different architectures will be realized.

[0079] For any size of task, the task can be divided in terms of timing, with the cycle (step) corresponding to the task being divided into high-level and the corresponding phase (phase) being divided into low-level timing. The phase (phase) is the minimum granularity of the handshake between two chips, and the cycle (step) is the maximum granularity of the handshake between two chips.

[0080] There are two requirements for the interaction between the two chips. The first is data exchange, and the second is control signal exchange. In the existing situation, the memristor chip is unit A and the silicon-based chip is unit B.

[0081] (1) After processing the data, A transmits it to B. B receives the complete data and starts the next operation.

[0082] (2) A processes and receives the complete data transmitted by B and starts the next operation

[0083] (3) A transmits a start signal to B, and B proceeds to the next step according to the content of the signal.

[0084] (4) A receives the complete signaling transmitted by B and proceeds to the next step according to the content.

[0085] According to the above, the complete data transmission and the complete signaling transmission are controlled as a single phase timing.

[0086] For example, a task is set up, and after splitting, different parts of the task are mapped to the memristor core and the silicon-based chip. There are a total of 6 complete data exchanges and 8 complete signaling controls between the two to ensure the orderliness of calculation and communication.

[0087] The tasks of the memristor core and silicon chip can be divided into ≥14 phases. If data exchange and signaling control transmission are intertwined, they can be divided into at least 8 phases. According to the above data, the handshake protocol process of a single data exchange and a single signaling exchange is as follows: Figure 5 As shown, there are the following descriptions:

[0088] (1) The memristor chip is selected as the master control unit of this task, and the signal is used to actively complete the cycle (step) start

[0089] (2) Enter the first phase to start, run, and end

[0090] (3) Data transmission

[0091] (4) Enter the second phase to start, run, and end

[0092] (5) Data transmission

[0093] (6) Run in sequence

[0094] There is a complete monitoring unit inside the circuit, which completes the cycle (step) implementation and phase (phase) implementation, involving the phase (phase) timing, operating status, corresponding to the task running process and other states.

[0095] This embodiment implements cross-modal circuit data encoding technology, solves the handshake problem of communication between the memristor storage and computing integrated core and the silicon-based accelerator core, and realizes efficient and stable data transmission between them. This embodiment can convert data signals of different modes into a unified encoding format for transmission and processing between the two cores. It not only helps to improve the performance of the entire system, but also promotes the application of memristor storage and computing integrated cores and silicon-based accelerator cores in a wider range of fields;

[0096] This embodiment organically integrates the two through on-chip interconnect communication networking, giving full play to their respective advantages, realizing digital-analog hybrid computing, and improving the performance of the overall system. The realized brain-like heterogeneous fusion chip design platform can realize brain-like general encoding and decoding technology and interconnection protocols, and can support multi-dimensional heterogeneous communication technology networks. Computing core particles of different forms such as adaptive brain-like core particles and control processor core particles have the ability to coordinate data flows in all dimensions, improve the flexibility of brain-like system construction and the accuracy of brain-like system construction, thereby building a larger-scale brain-like intelligent body and promoting the realization of brain-like intelligence;

[0097] This embodiment establishes system debugging and performance analysis through a handshake mechanism. System debugging and performance analysis can monitor and analyze the operating status of each chiplet and the overall system, so as to more intuitively understand and debug the communication, task scheduling and resource allocation between chiplets, and perform performance evaluation and optimization to ensure that it has good performance and energy efficiency in various computing tasks;

[0098] The on-chip communication interaction protocol between the memristor core and the silicon-based core of this embodiment needs to support the spatiotemporal coding technology that integrates the artificial neural network framework and the pulse neural network framework, explore heterogeneous data interaction technology, information loss and redundant coding schemes in the data interaction process, realize cross-modal data coding interaction communication packages, and support brain-like neural network algorithm applications. The multi-dimensional communication networking scheme of each heterogeneous unit realizes low-latency, high-speed communication control modules and physical communication interfaces, explores data retransmission and error correction technology with low latency, realizes flexible data transmission encoding and decoding technology, and supports high-speed data transmission of heterogeneous embedded brain-like computing.

[0099] In order to implement the above-mentioned embodiments, the present application also proposes a communication device in which a memristor core and a silicon-based core are interconnected and packaged.

[0100] Figure 6 A schematic diagram of the structure of the communication between a memristor core and a silicon-based core interconnect package provided in an embodiment of the present application.

[0101] like Figure 6 As shown, the communication device of the interconnected package of the memristive core and the silicon-based core comprises a memristive core and a silicon-based core, wherein the memristive core and the silicon-based core implement communication handshake according to the handshake protocol, the working timing of the memristive core and the silicon-based core is hierarchically divided according to the cycle (step) level and the phase (phase) level, the memristive core and the silicon-based core implement a cycle (step) timing control unit and a phase (phase) timing control unit inside, and the cycle (step) includes multiple phases (phase);

[0102] When a task is executed, in each cycle (step) stage, the memristive chip or silicon-based chip acts as the main control unit to start the cycle (step), and in each phase (phase) stage, the memristive chip or silicon-based chip completes the divided tasks and transmits control signals or data to complete the coordinated synchronization of interactive tasks, wherein the task is a task divided in task timing, and when dividing, the task corresponding to the cycle (step) is divided into a high-level timing, and the corresponding phase (phase) is divided into a low-level timing, the cycle (step) is the maximum granularity of the handshake between the chiplets, and the phase (phase) is the minimum granularity of the handshake between the chiplets.

[0103] Optionally, in one embodiment of the present application, the interaction requirements include data exchange and control signaling exchange, wherein the memristive core particle and the silicon-based core particle complete the data exchange, including:

[0104] In a single phase, the memristor chip completes the data processing and transmits the data to the silicon chip. The silicon chip receives the complete data and enters the next phase after the data transmission is completed. The silicon information starts the next operation, or,

[0105] In a single phase, the memristor chip processes the complete data transmitted by the silicon chip, and after the data transmission is completed, it enters the next phase, and the memristor chip starts the next operation.

[0106] The memristive core and the silicon-based core complete the control signaling exchange, including:

[0107] In a single phase, the memristor chip transmits control signals to the silicon chip, and after the transmission is completed, it enters the next phase, and the silicon chip proceeds to the next step according to the control signal content, or,

[0108] In a single phase timing, the memristor chip receives the control signal transmitted by the silicon-based chip, and after the transmission is completed, it enters the next phase timing, and the memristor chip proceeds to the next step according to the signal content.

[0109] Optionally, in one embodiment of the present application, the memristive core and the silicon-based core periodically perform data synchronization via signals within the chip, and the memristive core and the silicon-based core share data.

[0110] Optionally, in one embodiment of the present application, the control signaling in the handshake protocol includes: cycle (step)_start, cycle (step)_running, cycle (step)_progress, cycle (step)_finish, cycle (step)_ack, and the corresponding phase (phase) level includes phase (phase)_start, phase (phase)_running, phase (phase)_progress, phase (phase)_finish, phase (phase)_ack.

[0111] It should be noted that the aforementioned explanation of the communication method embodiment of the interconnection package of the memristive core and the silicon-based core is also applicable to the communication device of the interconnection package of the memristive core and the silicon-based core of this embodiment, and will not be repeated here.

[0112] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0113] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0114] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0115] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute the instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0116] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0117] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0118] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0119] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A communication method for interconnecting and packaging a memristive core particle and a silicon-based core particle, characterized in that: The memristive core particle and the silicon-based core particle implement a communication handshake according to a handshake protocol, and the method comprises the following steps: The working timing of memristive core particles and silicon-based core particles is hierarchically divided according to the cycle step level and phase level, and cycle step timing control and phase timing control are implemented inside the memristive core particles and silicon-based core particles, where the cycle step contains multiple phases; Get the tasks and divide them based on the task timing. The tasks corresponding to the loop step are divided into high-level timing, and the corresponding phase is divided into low-level timing. The loop step is the maximum granularity of the handshake between core particles, and the phase is the minimum granularity of the handshake between core particles. During task execution, at each loop step, the memristive core or silicon-based core acts as the master control unit to initiate the loop step. At each phase, the memristive core or silicon-based core completes the divided task and transmits control signals or data between the memristive core and silicon-based core to achieve coordinated synchronization of interactive tasks. The handshake protocol process of the loop step includes: During task execution, at each loop step, the memristive core particle or silicon-based core particle acts as the master control unit to send a loop step_start control signal to start the loop step, and the silicon-based core particle or memristive core particle receives the loop step_start control signal and feeds back the loop step_ack control signal. During task execution, at each loop step, the memristive core or silicon-based core sends a loop step_finish control signal to end the loop step, and the silicon-based core or memristive core receives the loop step_finish control signal and feeds back the loop step_ack control signal. The handshake protocol process of the phase phase includes: When the task is executed, in each loop step phase, after the loop step is started, each phase phase is entered in sequence; In each phase, the memristive core or silicon-based core sends a phase_start control signal to start the phase, and the silicon-based core or memristive core receives the phase_start control signal and feeds back the phase_ack control signal. In each phase phase, the memristive core or silicon-based core sends a phase phase_finish control signaling to end the phase phase, and the silicon-based core or memristive core receives the phase phase_finish control signaling and feeds back a phase phase_ack control signaling.

2. The method according to claim 1, wherein Interaction requirements include data exchange and control signaling exchange. The data exchange is completed by memristive core particles and silicon-based core particles, including: In a single phase, after the memristor core completes the data processing, it transmits the data to the silicon core. After the silicon core receives the complete data and completes the data transmission, it enters the next phase, and the silicon information initiates the next operation. Or, In a single phase, the memristive core processes the complete data transmitted by the silicon-based core. After the data transmission is completed, the next phase begins and the memristive core initiates the next operation. The control signaling exchange is completed by the memristive core and the silicon-based core, including: In a single phase, the control signal is transmitted to the silicon-based chip through the memristive chip, and after the transmission is completed, the next phase is entered, and the silicon-based chip enters the next operation according to the control signal content, or In a single phase sequence, the control signaling transmitted by the silicon-based core particle is received by the memristive core particle, and after the transmission is completed, the next phase sequence is entered, and the next operation is performed according to the signaling content by the memristive core particle.

3. The method according to claim 1, wherein Also includes: Data synchronization is performed regularly on the memristive core particle and the silicon-based core particle through signals within the chip, so that the memristive core particle and the silicon-based core particle share data.

4. The method according to claim 1, wherein The control signaling in the handshake protocol includes: cycle step_start, cycle step_running, cycle step_progress, cycle step_finish, cycle step_ack, and the corresponding phase level includes phase phase_start, phase phase_running, phase phase_progress, phase phase_finish, and phase phase_ack.

5. A communication device comprising a memristive core and a silicon-based core interconnected and packaged, characterized in that: include: Memristive core particles and silicon-based core particles, wherein the memristive core particles and the silicon-based core particles implement communication handshake according to the handshake protocol, the working timing of the memristive core particles and the silicon-based core particles is hierarchically divided according to the cycle step level and the phase level, the memristive core particles and the silicon-based core particles implement a cycle step timing control unit and a phase phase timing control unit inside, and the cycle step includes multiple phases; During task execution, at each cycle step, the memristive core particle or silicon-based core particle acts as the master control unit to start the cycle step, and at each phase phase, the memristive core particle or silicon-based core particle completes the divided task and transmits control signaling or data to complete the coordinated synchronization of interactive tasks. The tasks are divided in terms of task timing. During the division, the tasks corresponding to the cycle step are divided into high-level timing, and the corresponding phase phase is divided into low-level timing. The cycle step is the maximum granularity of the handshake between core particles, and the phase phase is the minimum granularity of the handshake between core particles. The handshake protocol process of the loop step includes: During task execution, at each loop step, the memristive core particle or silicon-based core particle acts as the master control unit to send a loop step_start control signal to start the loop step, and the silicon-based core particle or memristive core particle receives the loop step_start control signal and feeds back the loop step_ack control signal. During task execution, at each loop step, the memristive core or silicon-based core sends a loop step_finish control signal to end the loop step, and the silicon-based core or memristive core receives the loop step_finish control signal and feeds back the loop step_ack control signal. The handshake protocol process of the phase phase includes: When the task is executed, in each loop step phase, after the loop step is started, each phase phase is entered in sequence; In each phase, the memristive core or silicon-based core sends a phase_start control signal to start the phase, and the silicon-based core or memristive core receives the phase_start control signal and feeds back the phase_ack control signal. In each phase phase, the memristive core or silicon-based core sends a phase phase_finish control signaling to end the phase phase, and the silicon-based core or memristive core receives the phase phase_finish control signaling and feeds back a phase phase_ack control signaling.

6. The device according to claim 5, characterized in that Interaction requirements include data exchange and control signaling exchange. The memristive chip and silicon-based chip complete the data exchange, including: In a single phase, the memristor core completes data processing and transmits the data to the silicon core. The silicon core receives the complete data and enters the next phase after the data transmission is completed. The silicon core starts the next operation, or In a single phase, the memristive chip processes the complete data transmitted by the silicon chip. After the data transmission is completed, it enters the next phase and the memristive chip starts the next operation. The memristive core and the silicon-based core complete the control signaling exchange, including: In a single phase, the memristor core transmits control signals to the silicon core, and after the transmission is completed, it enters the next phase. The silicon core proceeds to the next step according to the control signal content, or, In a single phase timing sequence, the memristive chip receives the control signaling transmitted by the silicon-based chip, and after the transmission is completed, it enters the next phase timing sequence, and the memristive chip proceeds to the next step according to the signaling content.

7. The device according to claim 5, characterized in that The memristive core particle and the silicon-based core particle periodically perform data synchronization via signals within the chip, and the memristive core particle and the silicon-based core particle share data.

8. The device according to claim 5, wherein The control signaling in the handshake protocol includes: cycle step_start, cycle step_running, cycle step_progress, cycle step_finish, cycle step_ack, and the corresponding phase level includes phase phase_start, phase phase_running, phase phase_progress, phase phase_finish, and phase phase_ack.

Citation Information

Patent Citations

  • Low-power clocking for a high-speed memory interface

    CN107835988A

  • Storage and calculation integrated processor, processing system and algorithm model deployment method

    CN112836814A

  • Deep Learning-Based Polymorphic Platform

    US20220217035A1