White-Box GPU Confidential Computing Beyond the TEE Boundary
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in securely combining a Trusted Execution Environment (TEE) with a Graphics Processing Unit (GPU) for high-power confidential computing, as the GPU lacks a secure hardware root and existing methods fail to protect against unauthorized access, leading to data security risks during data transmission and computing.
Innovation Solution
The method involves scrambling encrypted data using a ciphertext scrambling function in a TEE and processing it with a target white-box decryption combined function on the GPU, ensuring data protection through ciphertext descrambling, decryption, and first data protection, and returning the result to the TEE for plaintext processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If encrypted data is transmitted from TEE to GPU for processing, then computing power is improved, but data security deteriorates due to lack of hardware root in GPU
Solution Approach 1:
The patent applies preliminary action by performing encryption and scrambling operations on data before it leaves the TEE. The data is encrypted using AES algorithm and then scrambled using a ciphertext scrambling function before being transmitted to the GPU, ensuring that even if the GPU is compromised, the original data remains protected.
Solution Approach 2:
The patent introduces an intermediary mechanism by using a trusted application layer that manages the encryption, scrambling, and unscrambling processes. This trusted application acts as a mediator between the TEE and GPU, controlling data flow and ensuring security protocols are followed without requiring the GPU to have inherent security capabilities.
2Productivity
If plaintext data is processed in GPU, then computing efficiency is improved, but data protection deteriorates making data vulnerable to attacks
Solution Approach 1:
The patent applies preliminary action by pre-encrypting and pre-scrambling data before it reaches the GPU. This ensures that the GPU never receives or processes plaintext data, eliminating the vulnerability window while maintaining computing efficiency through hardware acceleration of the scrambled ciphertext.
Solution Approach 2:
The patent converts the potential harm of GPU processing unsecured data into a benefit by designing the system so that the GPU processes scrambled ciphertext instead. The scrambling function appears to add complexity but actually provides security by ensuring that even if the GPU is attacked, the data remains unintelligible without the unscrambling key held by the trusted application.
3Reliability
If data is encrypted and scrambled before GPU processing, then data security is improved, but system complexity increases
Solution Approach 1:
The patent applies universality by designing the trusted application to handle multiple security functions (encryption, scrambling, unscrambling, key management) in a unified manner. The same trusted application framework is used throughout the system regardless of the specific computing task, reducing overall system complexity despite the added security layers.
Solution Approach 2:
The patent substitutes complex hardware security mechanisms with software-based security in the trusted application layer. Instead of requiring secure hardware roots in the GPU or complex hardware-trusted setups, the security is achieved through cryptographic algorithms and software-controlled access, simplifying the hardware requirements while maintaining strong security.
4Area of stationary object
If TEE security boundary is expanded to GPU, then security coverage is improved, but implementation difficulty increases
Solution Approach 1:
The patent uses an intermediary trusted application layer to extend the TEE security boundary to the GPU without requiring direct hardware integration. This intermediary handles all security-critical operations and communicates between the TEE and GPU, making the expansion of security coverage achievable through software rather than complex hardware modifications.
Solution Approach 2:
The patent replaces the need for physical hardware security extensions with a software-based trusted application framework. This substitution allows the security boundary to be expanded to the GPU through logical control and cryptographic protection rather than requiring physical security hardware in the GPU, significantly reducing implementation difficulty.
Data Source
Figure 1a~1b
Figure 2~3
Figure 4
AI summary
This specification provides methods and systems for implementing confidential computing, electronic devices, and storage media. The method can include: scrambling encrypted data based on a ciphertext scrambling function in a trusted execution environment, and sending scrambled encrypted data to a GPU, so that the GPU processes the scrambled encrypted data by using a target white-box decryption combined function, and obtains, through computing, a corresponding computing result by using obtained processed data, where the target white-box decryption combined function is used to perform ciphertext descrambling processing, decryption processing, and first data protection processing on the scrambled encrypted data; and obtaining the computing result or a derived result of the computing result, and performing processing in the trusted execution environment to obtain a plaintext computing result.