Neural Network Bit-Wise Correlation for Side Channel Key Recovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for training neural networks to enhance correlation coefficients in side channel attacks are limited in effectively utilizing profiling traces and loss functions to determine secret keys, particularly in cryptographic systems.
Innovation Solution
A method involving a neural network, such as a convolutional neural network, is trained using profiling traces of physical parameters like electrical power consumption, with a bit-wise correlation loss function and weighting coefficients to improve correlation coefficients, enabling the determination of secret keys in cryptographic systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional loss functions are used for training neural networks in side channel attacks, then training is simpler, but correlation coefficient is lower
Solution Approach 1:
The loss function is segmented into bit-wise components, where each bit of the intermediate value is processed separately. This segmentation allows the neural network to learn precise bit-level correlations from profiling traces, thereby increasing the correlation coefficient while maintaining manageable complexity through modular bit-level processing
Solution Approach 2:
The invention changes the parameter representation by using bit-wise decomposition of the intermediate values. Instead of treating intermediate values as continuous or whole units, the loss function operates on individual bits, transforming the parameter space to enable more precise correlation measurement and improve attack effectiveness
2Measurement precision
If more profiling traces are collected, then training accuracy improves, but measurement time increases
Solution Approach 1:
The invention extracts only the essential bit-level features from profiling traces that are most relevant to key recovery. By focusing on specific bit correlations rather than processing entire intermediate values, the method achieves high training accuracy with fewer traces, reducing measurement time while maintaining effectiveness
Solution Approach 2:
The bit-wise loss function applies partial action by selectively correlating specific bits of intermediate values with power consumption traces. This partial correlation approach achieves sufficient training accuracy without requiring complete analysis of all trace data, thereby reducing the time needed for data collection and processing
3Measurement precision
If bit-wise correlation loss function is used, then correlation coefficient increases, but computational complexity increases
Solution Approach 1:
The computational complexity is managed through segmentation of the loss function into independent bit-wise operations. Each bit correlation can be computed separately and efficiently, allowing parallel processing that reduces overall computational burden while maintaining high correlation coefficients through precise bit-level matching
Solution Approach 2:
By changing to bit-wise parameter representation, the invention transforms complex continuous optimization into discrete bit-level correlation tasks. This parameter transformation simplifies the computational structure despite the increased precision, as bit-wise operations are computationally more efficient than continuous value matching
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 bit-wise correlation of an output value provided by the neural network and a predetermined function characterizing an operation of a physical system.