Neural Network Circuit Trojan Detection via Signal Summary
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Digital circuits may contain design defects, including malicious code, that are difficult to detect as they can remain dormant and only manifest under specific conditions, posing a risk to sensitive systems.
Innovation Solution
A system and method using a discriminative neural network to analyze a value change summary file generated from simulating a digital circuit design, trained on a corpus of designs with and without intermittent severe defects, to detect such defects by compiling signal change data and applying dimensionality reduction and data compression techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional detection methods are used, then the system is simpler, but the ability to detect intermittent severe defects is insufficient
Solution Approach 1:
The patent replaces traditional mechanical/deterministic detection methods with a neural network-based system that uses learned patterns from training data to detect intermittent severe defects, enabling more reliable detection of subtle and rare defect patterns
Solution Approach 2:
The patent creates a virtual copy of the digital circuit design through simulation to generate training data and test cases, allowing the neural network to learn from replicated circuit behaviors without modifying the actual physical circuit
2Measurement precision
If more training data is used, then the neural network accuracy improves, but the data processing time increases
Solution Approach 1:
The patent performs preliminary actions by pre-training the neural network on a comprehensive dataset of circuit designs with and without defects before actual detection is needed, allowing the model to make accurate predictions quickly during deployment
Solution Approach 2:
The patent uses excessive training data to ensure the neural network captures all possible defect patterns, then leverages the pre-trained model's knowledge to perform rapid inference on new circuits without reprocessing the entire training dataset
3Reliability
If simulation time is increased, then the detection accuracy improves, but the computational resources consumed increase
Solution Approach 1:
The patent extracts only the critical information needed for defect detection from the full simulation output, focusing the neural network analysis on key signal patterns and behavioral characteristics rather than processing every simulation detail
Solution Approach 2:
The patent performs preliminary simulation and feature extraction during the training phase to build a comprehensive model of normal and defective circuit behaviors, allowing rapid classification of new circuits without repeating the full simulation process
Data Source
AI summary
A system and method for detecting Trojans and other intermittent severe defects in a digital circuit design. A simulation of the digital circuit design results in a value change dump file, which is compiled to form a value change summary file containing counts of the numbers of value changes for the signals in the digital circuit design. A discriminative neural network analyzes the value change summary file to determine whether an intermittent severe defect is present. A corpus of digital circuit designs, with and without intermittent severe defects, is used to train the discriminative neural network. The training process may involve dimensionality reduction of the data, enlargement of the data set, and data compression using an autoencoder.