Floating-Point Masking for Side Channel Attack Resistance

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing masking countermeasures against side channel attacks, particularly in cryptography, are vulnerable to deep machine learning approaches and high-order attacks, which can reveal masked data and exploit leakage.

Innovation Solution

The method involves using floating-point operations in conjunction with an invertible function that introduces non-null rounding errors, generating output binary words with a large Hamming distance, thereby decorrelating the distance of input integer numbers and output binary representations, making it harder for deep learning machines to identify masked data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If affine masking countermeasures are used to protect substitution tables against side channel attacks, then security against traditional attacks is improved, but vulnerability to deep machine learning attacks and high-order attacks increases

Engineering Contradiction:

Improvesecurity against side channel attacksVSAvoidvulnerability to deep learning attacks

Core Design Contradiction:

ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent transforms the masking operation from simple XOR-based Boolean masking to floating-point multiplication with invertible functions. This parameter change in the mathematical domain creates non-linear transformations that decorrelate input-output relationships, making deep learning attacks ineffective while maintaining resistance against traditional side channel and high-order attacks.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent introduces floating-point arithmetic operations as an intermediary layer between the input data and the substitution table access. This intermediary transformation layer with invertible functions and controlled rounding errors breaks the direct correlation that deep learning models exploit, while still allowing legitimate cryptographic operations to proceed.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Object-affected harmful factors

If floating-point operations with invertible functions are used to generate output binary words, then resistance against deep learning attacks is improved, but computational complexity increases

Engineering Contradiction:

Improveresistance to deep learning attacksVSAvoidcomputational complexity

Core Design Contradiction:

Object-affected harmful factorsVSDevice complexity

Solution Approach 1:

The patent pre-defines specific invertible floating-point functions and rounding error characteristics that will be used in the masking operation. By establishing these parameters in advance, the system avoids the need for complex dynamic function selection during execution, reducing runtime computational complexity while maintaining the security benefits of the floating-point transformation approach.

Inventive Principle:

Principle #10Preliminary action

3Difficulty of detecting and measuring

If floating-point multiplication and invertible functions are applied to mask data, then Hamming distance between output binary words increases, but energy consumption increases

Engineering Contradiction:

ImproveHamming distance decorrelationVSAvoidenergy consumption

Core Design Contradiction:

Difficulty of detecting and measuringVSUse of energy by stationary object

Solution Approach 1:

The patent applies floating-point operations selectively to specific critical data paths where masking is most needed, rather than uniformly applying complex transformations to all data. This localized application of floating-point multiplication and invertible functions achieves sufficient decorrelation for security while minimizing unnecessary energy consumption in less critical operations.

Inventive Principle:

Principle #3Local quality

Data Source

PatentEP4542372A1Method of generating an output binary word from an input binary word, method of performing an arithmetic operation on an input binary word associated to an output binary word, corresponding computer program product and devices

Publication Date: 2025.04.23 NAGRAVISION SA

Figure 1~2
Figure 3~3a
Figure 4~5

AI summary

A method is proposed for the generation of an output binary word from an input binary word. Such method comprises: - executing a first floating-point multiplication (5200) wherein the input binary word interpreted as an integer number is multiplied with a first predetermined number of floating-point type, resulting in a first argument number of floating-point type; - applying a predetermined invertible function (5220) to the first argument number delivering a first output number of the floating-point type, the output binary word being based on the first output number. The predetermined invertible function is configured to have the first output number having a non-null output rounding error resulting in having an application, to the first output number, of an invert function of the predetermined invertible function that delivers an input number of the floating-point type corresponding to the input operand with a non-null input error.