Artificial intelligence processor security enhancement system and method based on key path encryption

Abstract

The invention relates to an artificial intelligence processor security enhancement system and method based on key path encryption. The artificial intelligence processor security enhancement system iscomposed of an instruction security unit, a data security unit and a universal artificial intelligence processor architecture. On the basis of a universal artificial intelligence processor architecture, an instruction security unit and a data security unit are added to ensure the security of instructions and weights of a neural network model and protect the security of intermediate data processedby an artificial intelligence processor. According to the technology, a private key is transmitted through an asymmetric encryption algorithm to decrypt an encrypted instruction and a weight, the integrity of a neural network model instruction can be verified through a bypass chain type verification method, and the instruction transmission performance of an artificial intelligence processor is notaffected. The system adopts encryption algorithms (e.g., advanced encryption algorithms such as AES-CTR mode) to encrypte a feature map of a critical path, so that the privacy of the feature map is protected, side channel information leakage of an artificial intelligence processor is reduced, and an attacker cannot infer the structure of the neural network model through an access mode of an off-chip DRAM. The method has wide practical value and application prospect.

Description

technical field [0001] The invention relates to an artificial intelligence processor security enhancement system and method, including model security, data security, and instruction security of the artificial intelligence processor. It is mainly used to protect the security of the model, data and instructions run by the AI ​​processor. The artificial intelligence processor security enhancement technology mainly uses encryption algorithms (such as AES encryption algorithm) to encrypt and decrypt model weights, instructions, and intermediate data processed by artificial intelligence processors to ensure the security of artificial intelligence processors. It belongs to computing systems and The field of microprocessor security. Background technique [0002] As the most energy-efficient artificial intelligence processor, ASIC has attracted extensive attention from industry and academia. In academia, in 2014, Chen Yunji's team from the Institute of Computing Technology, Chinese...

AI Technical Summary

Problems solved by technology

[0005] The technical problem of the present invention is to overcome the deficiencies of the prior art, and provide an artificial intelligence processor security enhancement system and method based on critical path encryption, which can effectively decrypt the weight and operation instructions, encryption and decryption of the neural network model from the CPU encryption The intermediate data processed by the artificial intelligence processor makes up for the defect that the existing artificial intelligence processor is vulnerable to side channel attacks and ensures the safety of the artificial intelligence processor; it has novel structure, small size, high performance, strong encryption performance and adaptability Strong and other characteristics

Images