Task definition type storage-computing integrated artificial intelligence computing device, method and system

Abstract

The application discloses a task definition type storage-computation integrated artificial intelligence computing device, method and system. The artificial intelligence computing device comprises at least one storage-computation integrated unit and configuration control logic, the storage-computation integrated unit can be configured according to a predefined task structure, and directly completes an artificial intelligence computing task without an operating system and general instruction execution. The task structure is used for defining the computing behavior of the storage-computation integrated unit, so that the artificial intelligence computing process is determined by the task structure itself instead of being driven by a general instruction flow. The artificial intelligence computing device of the application can be embedded into various electronic devices or systems as an independent artificial intelligence computing element, and is suitable for application scenarios such as Internet of Things, edge computing and distributed artificial intelligence.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence computing technology, specifically to an integrated storage-computing artificial intelligence computing device, method, and system, and particularly to an artificial intelligence computing paradigm that directly defines computing behavior through task structure. Background Technology

[0002] With the widespread application of artificial intelligence technology in fields such as image recognition, speech recognition, natural language processing, intelligent sensing, and automatic control, traditional computing methods based on the von Neumann architecture face "memory walls" and power consumption bottlenecks when performing artificial intelligence tasks: the separation of memory and calculator leads to frequent transfer of large amounts of data between the two, resulting in high latency and high energy consumption; at the same time, existing artificial intelligence computing relies heavily on operating systems, drivers, and general instruction sets to complete model loading, data scheduling, and computation execution, making the system complex, costly, and difficult to embed into low-cost terminal devices.

[0003] In-memory computing technologies that have emerged in recent years have alleviated data transfer problems to some extent, but most existing solutions still retain an instruction-based computing paradigm, relying on software stacks and system scheduling, making them unsuitable as simple, low-cost, and scalable basic computing components. Therefore, proposing a new computing paradigm that allows computing behavior to be directly defined by task structure, thereby enabling the completion of artificial intelligence tasks without the need for operating systems and general instruction execution, has significant practical value and industrial implications. Summary of the Invention

[0004] (a) Purpose of the invention The purpose of this invention is to provide a task-defined storage-computing integrated artificial intelligence computing device, method and system, which aims to directly define computing behavior with task structure, thereby reducing system complexity, improving energy efficiency, and enabling artificial intelligence computing capabilities to be embedded as independent, low-cost computing components in various terminals and distributed systems.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a task-defined storage-computing integrated artificial intelligence computing device, comprising at least one storage-computing integrated unit and configuration control logic; the storage-computing integrated unit can be configured according to a predefined task structure to complete artificial intelligence computing tasks without the need for an operating system and general instruction execution. The task structure is used to define the computing behavior of the storage-computing integrated unit, so that after input data enters the storage-computing integrated unit, computing results are directly generated according to the task structure. The task structure can exist in the form of a set of weight parameters, connection / topology relationships, data flow control logic, etc., and can be loaded, switched, or solidified during manufacturing.

[0006] The present invention also provides a corresponding method and system. The method is characterized in that, without the need for an operating system and the execution of general instructions, it completes artificial intelligence computing tasks by having at least one storage-computing unit perform operations according to a predefined task structure. The system is characterized in that it includes at least one artificial intelligence computing device as described above, for completing artificial intelligence computing tasks locally in the system and outputting computing results.

[0007] (III) Beneficial Effects Compared with existing technologies, the present invention has at least the following beneficial effects: by directly defining computing behavior through task structure, it eliminates the dependence on operating system and general instruction flow, significantly reducing system complexity and energy consumption; the use of in-memory computing unit can reduce data handling and improve computing energy efficiency; the device of the present invention can be packaged as an independent component, which is convenient for large-scale deployment in IoT, edge computing and distributed systems, and can achieve flexible configuration or specialization of computing functions by loading or fixing task structure, thereby adapting to different application scenarios and deployment modes. Attached Figure Description

[0008] • Figure 1 A schematic diagram of a task-defined storage-computing integrated artificial intelligence computing device; • Figure 2 A schematic diagram illustrating how a storage-computing unit receives input data and outputs computation results; • Figure 3 This is a schematic diagram of the structure of an artificial intelligence computing system based on the artificial intelligence computing device of the present invention. Detailed Implementation

[0009] The following examples are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0010] Example 1: Artificial intelligence computing device with reconfigurable task structure (see...) Figure 1 ) like Figure 1 As shown, the artificial intelligence computing device in this embodiment includes configuration control logic and a storage-computing integrated unit. The configuration control logic sends configuration instructions or parameters to the storage-computing integrated unit through a control interface, causing the storage-computing integrated unit to operate according to the parameter / data flow relationship defined by the task structure. The task structure can be in the form of a set of weight parameters and data flow control logic, stored in a task structure storage module or loaded online through the configuration control logic. In this embodiment, after receiving an input signal, the storage-computing integrated unit does not require an operating system or general instruction calls, but directly performs parallel calculations such as multiplication-addition and merging at the hardware / hybrid analog-digital level and outputs the results.

[0011] In some implementations, the configuration control logic is configured to load new task structures from external storage or other parts of the system and write them to the task structure storage module to achieve model replacement or functional upgrade of the artificial intelligence computing device.

[0012] Example 2: Dedicated artificial intelligence computing device with fixed task structure (see Figure 2 ) like Figure 2 As shown, input data enters the storage-computing unit through a data path. The storage-computing unit performs calculations based on the task structure fixed during manufacturing and directly outputs the calculation results. This embodiment is suitable for terminal application scenarios with fixed functional requirements and cost sensitivity, such as fixed keyword wake-up, specific signal recognition, or simple sensor data preprocessing. The fixed task structure can be set in a one-time write-through during manufacturing, thereby eliminating the need for a runtime loading mechanism and further reducing device complexity and cost.

[0013] This embodiment emphasizes: Figure 2 The terms "input data" and "output results" are placed outside the boxes as part of the data flow relationship, indicating that they are not essential physical components of the device, but rather external data entities that interact with the device. The description and limitations of the task structure are presented in the text of the specification and claims, rather than within the illustrated boxes, to avoid limiting the illustration.

[0014] Example 3: Artificial intelligence computing systems deployed at the edge / Internet of Things (see Figure 3 ) like Figure 3 As shown, the artificial intelligence computing system includes at least one artificial intelligence computing device. In this embodiment, the artificial intelligence computing device is deployed as an independent computing node in the system, performing judgments or inferences locally on sensor data or local information. The system only receives the calculation results output by the device, without uploading the original data, thus reducing communication bandwidth and privacy exposure risks. This system can be an edge computing system or a widely distributed Internet of Things (IoT) system, and specific applications may include intelligent monitoring, real-time control, and privacy-preserving local inference.

[0015] Other embodiments and variations The task structure of this invention can be stored internally in the device or partially managed in an external system and deployed when needed. The storage-computing integrated unit can be implemented using memristor arrays, resistive switching memory arrays, analog / mixed-signal parallel arrays, or other physical structures capable of achieving in-memory computing functionality. This invention is not limited to any specific material, process, or interface protocol; any equivalent substitutions that do not depart from the technical essence of this invention are within its protection scope.

Claims

1. A task-defined storage-computing integrated artificial intelligence computing device, characterized in that, include: At least one storage-compute integrated unit; And configuration control logic; wherein, the storage-computing integrated unit can be configured according to a predefined task structure to complete artificial intelligence computing tasks without the need for an operating system and general instruction execution.

2. The artificial intelligence computing device according to claim 1, characterized in that, The task structure is used to define the computational behavior of the storage-computing unit, so that it can directly generate corresponding computational results when receiving input data.

3. The artificial intelligence computing device according to claim 1, characterized in that, The task structure is embodied in a set of weight parameters, connection relationships, and / or data flow control logic.

4. The artificial intelligence computing device according to claim 1, characterized in that, It also includes a task structure storage module for storing the task structure.

5. The artificial intelligence computing device according to claim 1, characterized in that, The configuration control logic is used to select, load, switch, or solidify different task structures so that the same storage-computing unit can adapt to different artificial intelligence computing tasks.

6. The artificial intelligence computing device according to claim 1, characterized in that, The storage-computing integrated unit is a storage-computing integrated structure, in which storage and computing functions are physically or logically integrated.

7. The artificial intelligence computing device according to claim 1, characterized in that, The artificial intelligence computing device is packaged as a standalone artificial intelligence computing element and embedded as a standard electronic device into other electronic devices or systems.

8. A task-defined storage-computing integrated artificial intelligence computing method, characterized in that: Artificial intelligence computing tasks can be completed by having at least one storage-computing unit perform operations according to a predefined task structure without the need for an operating system and general instruction execution.

9. The artificial intelligence calculation method according to claim 8, characterized in that, The task structure is applied to the storage-computing unit by loading, switching, or fixing.

10. An artificial intelligence computing system, characterized in that, It includes at least one artificial intelligence computing device according to any one of claims 1 to 7, for performing artificial intelligence computing tasks locally in the system.

11. The artificial intelligence computing system according to claim 10, characterized in that, The artificial intelligence computing device performs artificial intelligence computing tasks locally within the system and outputs the computing results.

12. The artificial intelligence computing system according to claim 10, characterized in that, The system is an edge computing system or an Internet of Things (IoT) system.

13. The artificial intelligence computing system according to claim 10, characterized in that, The system only outputs the results of artificial intelligence calculations and does not output the raw sensor data.

Images