Neural Network Cryptography Coprocessor Against Side-Channel Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing countermeasures against sophisticated side-channel analysis attacks, particularly those using convolutional neural networks, are inadequate in protecting security devices from revealing sensitive information.
Innovation Solution
A method involving training a protective neural network on a coprocessor to select a combination of elementary protections based on the sensitive information being processed, using a training data set generated by applying various protection combinations and observing side-channel data to determine effective countermeasures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If neural network operations are implemented in software on a general-purpose processor, then flexibility and adaptability are improved, but security against side-channel analysis deteriorates
Solution Approach 1:
The patent introduces a specialized neural network cryptography coprocessor as an intermediary hardware component that handles neural network operations. This coprocessor acts as a mediator between the software layer and the main processor, providing hardware-accelerated neural network computations while maintaining security through dedicated cryptographic protection mechanisms, thus resolving the contradiction between software flexibility and hardware security.
Solution Approach 2:
The patent replaces software-based neural network operations on general-purpose processors with hardware-based neural network operations on a specialized coprocessor. This substitution of mechanical/software implementation with dedicated hardware implementation provides both the flexibility of programmable neural networks and the security of hardware isolation, protecting against side-channel analysis.
2Object-affected harmful factors
If neural network operations are implemented in hardware on a specialized coprocessor, then security against side-channel analysis is improved, but device complexity increases
Solution Approach 1:
The patent designs a universal neural network cryptography coprocessor that can perform multiple neural network operations (convolutions, activations, pooling, normalization) and support various neural network architectures. This multi-functional design reduces overall system complexity by consolidating diverse neural network functions into a single specialized hardware unit, rather than requiring separate hardware components for each operation.
Solution Approach 2:
The patent segments the neural network processing functionality into distinct operational units within the coprocessor (e.g., separate units for convolution, activation, pooling, normalization). This segmentation allows each unit to be optimized for its specific function while maintaining a modular architecture that manages hardware complexity through organized functional decomposition.
3Ease of manufacture
If standard hardware architectures are used for neural network processing, then ease of manufacture is improved, but resistance to side-channel analysis deteriorates
Solution Approach 1:
The patent introduces a specialized neural network cryptography coprocessor as an intermediary hardware component that handles neural network operations. This coprocessor acts as a mediator between the software layer and the main processor, providing hardware-accelerated neural network computations while maintaining security through dedicated cryptographic protection mechanisms, thus resolving the contradiction between software flexibility and hardware security.
4Object-affected harmful factors
If countermeasures against side-channel analysis are implemented, then security is improved, but computational overhead and processing time increase
Solution Approach 1:
The patent implements countermeasures against side-channel analysis at the hardware design stage rather than adding them as software layers during operation. By embedding cryptographic protection mechanisms and secure arithmetic units directly into the coprocessor architecture from the beginning, the system achieves security without runtime overhead, as the protective measures are inherently part of the computational fabric.
Solution Approach 2:
The patent designs the coprocessor to inherently provide its own security against side-channel analysis through built-in cryptographic protection mechanisms and secure arithmetic operations. The system serves its own security needs through self-contained hardware-based protection, eliminating the need for additional computational overhead from external security layers or software-based countermeasures.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates to a method for securing a security device against side-channel analysis attacks while performing a sensitive operation, the method comprising: • training (401) an attack neural network to perform a side-channel attack against the security device while performing a sensitive operation; • creating (403) a training data set for a protective neural network by applying a plurality of elementary protection combinations to the sensitive operation while performing the sensitive operation using a plurality of values for the piece of sensitive information, and for each elementary protection combination and sensitive information value, recording in the training data set whether the elementary protection combination prevented the attack neural network from discerning the sensitive information value; • training (405) a protective neural network executing on a coprocessor of the security device using the training data set for the protective neural network such that an input to the protective neural network is a sensitive information value to be protected and an output of the protective neural network is an indicator of which combination of elementary protections to apply to protect the piece of information from being detectable using the attack neural network thereby producing a set of parameters for the protective neural network; and • Programming (407) the coprocessor of the security device with the set of parameters for the protective neural network.