Neural Network Weighting for Side-Channel Attack Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural network methods are ineffective in enhancing correlation coefficients for side-channel attacks, particularly in determining secret keys from physical systems, as they fail to adequately account for bit-wise and non-bit-wise correlations and hardware-specific parameters like parasitic capacitance.
Innovation Solution
A method involving a neural network that weights bit values using coefficients based on hardware aspects, trains on profiling traces of physical systems, and uses additional neural networks to approximate these coefficients, thereby improving correlation analysis and reducing asymmetry in bit lines to harden systems against side-channel attacks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If standard neural network training methods are used, then the neural network can be trained on profiling traces, but the correlation coefficient for side-channel attacks remains insufficient
Solution Approach 1:
The patent segments the neural network into multiple specialized components: a main neural network for processing profiling traces, additional neural networks for approximating weighting coefficients, and separate processing paths for different bit values. This segmentation allows each component to be optimized for its specific function, thereby improving the overall correlation coefficient without overwhelming complexity
Solution Approach 2:
The patent introduces weighting coefficients as an additional dimension in the neural network processing. Instead of treating all bit values equally, the system adds a weighting dimension that emphasizes certain bits over others based on their importance for side-channel attacks. This dimensional enhancement improves measurement precision by focusing computational resources on the most informative features
2Measurement precision
If bit-wise correlation is evaluated, then more detailed analysis is possible, but computational complexity increases
Solution Approach 1:
The patent applies local quality by treating different bit values differently through weighting coefficients. Instead of uniform processing, each bit position can be weighted according to its local importance in revealing secret key information. This allows high-precision bit-wise correlation where needed while avoiding unnecessary computational expenditure on less informative bits
Solution Approach 2:
The system performs partial action by focusing computational effort on the most relevant bit values through weighting. Rather than exhaustively analyzing all possible correlations with equal depth, the neural network concentrates resources on bits that provide the most information for side-channel attacks, achieving high precision with reduced computational power
3Measurement precision
If weighting coefficients are introduced for bit values, then correlation analysis improves, but the system requires additional neural networks and training data
Solution Approach 1:
The additional neural networks introduced for weighting coefficient approximation are designed to be universal components that can be integrated with any main neural network for side-channel analysis. These weighting networks serve multiple purposes: they approximate coefficients for different bit positions, adapt to different cryptographic functions, and can be trained on various types of profiling traces. This multi-functionality justifies the increased device complexity by providing a flexible framework that improves correlation analysis across diverse applications
Data Source
AI summary
A computer-implemented method for a neural network, for example an artificial deep neural network. The method includes: providing a plurality of training data sets, each training data set comprising input data for the neural network and associated output data, training the neural network based on the plurality of training data sets and a loss function, wherein the loss function is based on a correlation of an output value provided by the neural network and a predetermined function characterizing an operation of a physical system, wherein the method further comprises weighting bit values of a leakage value associated with the predetermined function using weighting coefficients, wherein weighted bit values are obtained, and evaluating the correlation based on the weighted bit values.