Neural Network Segmentation Secure Element
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks in mobile devices are vulnerable to attacks and tampering, especially when the network structure and weights are stored in memory, and the communication between processing systems adds a point of vulnerability, making them susceptible to cloning and manipulation.
Innovation Solution
The solution involves dividing a neural network into two portions, with a first processing system handling the input layer and a secure element handling the hidden layers, ensuring that the neural network model is stored and executed within a secure element, preventing unauthorized access and tampering by limiting the number of executions through a velocity mechanism.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If neural network model is stored in memory of processing system, then ease of operation is improved, but security is worsened due to vulnerability to attacks and tampering
Solution Approach 1:
The neural network model is divided into two portions: a first portion stored in the memory of the first processing system and a second portion stored in a secure element. This segmentation allows the system to maintain ease of operation for the first portion while protecting the critical second portion in a tamper-resistant environment, thus resolving the contradiction between operational ease and security.
Solution Approach 2:
A secure element is introduced as an intermediary storage component between the processing system and the neural network model. This secure element acts as a mediator that provides tamper-resistant storage for the second portion of the model, enabling secure access without compromising the overall system operation. The secure element communicates with the first processing system through defined interfaces, maintaining operational flow while enhancing security.
2Adaptability or versatility
If communication between processing systems is enabled, then functionality is improved, but security is worsened due to additional vulnerability points
Solution Approach 1:
The second portion of the neural network model is extracted from the general-purpose memory system and placed into a dedicated secure element. This extraction removes the most critical data from the vulnerable communication path between processing systems, reducing the attack surface while maintaining the necessary functionality through controlled access to the secure element.
3Productivity
If neural network is executed on external processing system, then computational performance is improved, but security is worsened due to exposure of model
Solution Approach 1:
Different portions of the neural network model are stored with different security characteristics: the first portion in standard memory accessible by the first processing system, and the second portion in a tamper-resistant secure element. This local differentiation of security quality allows computational performance to be optimized by executing on external systems while critical model portions remain protected in the secure element, preventing unauthorized exposure.
Data Source
AI summary
An embodiment apparatus comprises a first processing system executing a first portion of a neural network comprising a first subset of a set of neural network layers providing a first intermediate output, and a second processing system receiving the first intermediate output, and operating a second portion of the neural network comprising a second subset of the set of layers providing a respective output, the second processing system configured to supply to the first processing system an output information function of the respective output, and the first processing system configured to obtain as a function of the output information a final output of the neural network. The second processing system includes a secure element storing a model of the second portion, and executes the second portion by applying the input information to the model of the second portion to provide the respective output.