Artificial intelligence AI processing method and AI processing apparatus

Patent metadata

Field	Value
Publication number	US-11954204-B2
Application number	US-202117172804-A
Country	US
Kind code	B2
Filing date	Feb 10, 2021
Priority date	Aug 14, 2018
Publication date	Apr 9, 2024
Grant date	Apr 9, 2024

Abstract

Embodiments of the present invention disclose an AI processing method and an AI processing apparatus. The method is applied to the AI processing apparatus. An AI processor has at least two working modes, and security of the at least two working modes is different. The method includes: processing, by the AI processor, an AI processing request in a target mode. The target mode is one of the at least two working modes, and the target mode is a working mode determined based on the AI processing request. The AI processor has at least two working modes with different security, and may switch between different working modes.

First claim

What is claimed is:

1. A method, comprising:
receiving, by a central processing unit (CPU) of an artificial intelligence (AI) processing apparatus, an AI processing request, wherein the AI processing apparatus comprises a hardware layer including an AI processor and the CPU, wherein the AI processor is a neural processing unit (NPU) or a graphics processing unit (GPU), wherein each of the CPU and the AI processor has at least two working modes and can switch between the at least two working modes, and wherein security of the at least two working modes is different;
determining, by the CPU, a target mode corresponding to the AI processing request, wherein the target mode is one of the at least two working modes, wherein the AI processing request includes information for AI processing, and wherein the target mode of the AI processor is determined based on the information for AI processing corresponding to the target mode;
executing, by the CPU, a security policy corresponding to the target mode; and
processing, by the AI processor, the AI processing request in the target mode.

2. The method according to claim 1 , wherein the AI processing apparatus further comprises a controller connected to the AI processor, and the method further comprises:
before the processing the AI processing request in the target mode:
receiving, by the controller, the AI processing request; and
determining, by the controller, the target mode corresponding to the AI processing request.

3. The method according to claim 2 , wherein the determining, by the controller, the target mode corresponding to the AI processing request comprises:
determining, by the controller based on a service type requested by the AI processing request, the target mode corresponding to the AI processing request.

4. The method according to claim 2 , wherein the determining, by the controller, the target mode corresponding to the AI processing request comprises:
determining, by the controller based on a type of service data carried in the AI processing request, the target mode corresponding to the AI processing request.

5. The method according to claim 2 , wherein the method further comprises:
before the processing the AI processing request in the target mode:
setting, by the controller, a working mode of the AI processor to the target mode.

6. The method according to claim 2 , wherein the AI processing apparatus supports a rich execution environment (REE) and a trusted execution environment (TEE), and the method further comprises:
when the target mode is a first mode, initiating, by the controller, the AI processing request to the AI processor by using a first driver in the TEE; and
when the target mode is a second mode, initiating, by the controller, the AI processing request to the AI processor by using a second driver in the REE, wherein security of the AI processor in the first mode is higher than security of the AI processor in the second mode.

7. The method according to claim 6 , wherein the method further comprises:
before the initiating the AI processing request to the AI processor by using the first driver in the TEE:
when a running environment of the controller is the REE, switching, by the controller, the running environment from the REE to the TEE.

8. The method according to claim 6 , wherein the method further comprises:
before the initiating, by the controller, the AI processing request to the AI processor by using the second driver in the REE:
when a running environment of the controller is the TEE, switching, by the controller, the running environment from the TEE to the REE.

9. The method according to claim 2 , wherein the method further comprises:
before the processing, by the AI processor, the AI processing request in the target mode:
applying, by the CPU, for storage space required to process the AI processing request; and

setting, by the CPU, access permission for the storage space as target access permission by using a memory controller, wherein the target access permission is determined based on a service type requested by the AI processing request,
wherein the processing the AI processing request in the target mode comprises:
processing, by the AI processor, the AI processing request in the target mode by using the storage space.

10. The method of claim 1 , wherein the at least two working modes include three working modes, and wherein the AI processor processes the AI processing request in the target mode by using a corresponding AI model and a corresponding AI operator library code for the target mode.

11. An artificial intelligence (AI) processing apparatus, comprising:
a hardware layer including a hardware AI processor and a central processing unit (CPU), wherein the hardware AI processor is a neural processing unit (NPU) or a graphics processing unit (GPU), wherein each of the CPU and the hardware AI processor has at least two working modes and can switch between the at least two working modes, and wherein security of the at least two working modes is different,
wherein the CPU is configured to:
receive an AI processing request;
determine a target mode corresponding to the AI processing request, wherein the target mode is one of the at least two working modes, wherein the AI processing request includes information for AI processing, and wherein the target mode of the AI processor is determined based on the information for AI processing corresponding to the target mode; and
execute a security policy corresponding to the target mode, and

wherein the hardware AI processor is configured to:
process the AI processing request in the target mode.

12. The AI processing apparatus according to claim 11 , wherein the AI processing apparatus further comprises:
a controller connected to the hardware AI processor, configured to:
receive the AI processing request, and
determine the target mode corresponding to the AI processing request.

13. The AI processing apparatus according to claim 12 , wherein the controller is configured to:
determine, based on a service type requested by the AI processing request, the target mode corresponding to the AI processing request.

14. The AI processing apparatus according to claim 12 , wherein the controller is configured to:
determine, based on a type of service data carried in the AI processing request, the target mode corresponding to the AI processing request.

15. The AI processing apparatus according to claim 12 , wherein the controller is further configured to:
set a working mode of the hardware AI processor to the target mode.

16. The AI processing apparatus according to claim 12 , wherein the AI processing apparatus supports a rich execution environment (REE) and a trusted execution environment (TEE), and the controller is configured to:
when the target mode is a first mode, initiate the AI processing request to the hardware AI processor by using a first driver in the TEE, and
when the target mode is a second mode, initiate the AI processing request to the hardware AI processor by using a second driver in the REE, wherein security of the hardware AI processor in the first mode is higher than security of the hardware AI processor in the second mode.

17. The AI processing apparatus according to claim 16 , wherein the controller is further configured to:
when a running environment of the controller is the REE, switch the running environment from the REE to the TEE.

18. The AI processing apparatus according to claim 16 , wherein the controller is further configured to:
when a running enviro

Assignees

• Huawei Tech Co Ltd

Inventors

• Zhang Jianyong

Classifications

• G06F21/57 Primary

  Certifying or maintaining trusted computer platforms, e.g. secure boots or power-downs, version controls, system software checks, secure updates or assessing vulnerabilities · CPC title

• G06F9/4401

  Bootstrapping (security arrangements therefor G06F21/57) · CPC title

• G06F9/4411

  Configuring for operating with peripheral devices; Loading of device drivers · CPC title

• G06F9/445

  Program loading or initiating (bootstrapping G06F9/4401; security arrangements for program loading or initiating G06F21/57) · CPC title

• G06N5/04

  Inference or reasoning models · CPC title