NoC Router HT Detection and Dynamic Isolation Using Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network-on-chip (NoC) architectures are vulnerable to Hardware Trojans (HTs), which can cause performance degradation and security threats, and conventional detection and isolation techniques often result in false detections, power consumption increases, and network latency.
Innovation Solution
A machine learning-based methodology is employed to enhance HT detection and mitigation in NoC architectures, utilizing an artificial neural network for per-router HT detection and deep reinforcement learning for dynamic HT isolation, thereby optimizing performance, energy efficiency, and security.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional detection techniques (fault history logging, runtime threshold monitoring) are used to detect Hardware Trojans, then HT detection capability is provided, but false detection and misdetection occur leading to increased power consumption and network latency
Solution Approach 1:
The patent replaces conventional mechanical/threshold-based detection systems with a machine learning-based detection system. The ML model analyzes multiple NoC attributes (buffer utilization, link utilization, temperature, power consumption) to detect HTs, substituting the simple threshold monitoring mechanism with an intelligent system that reduces false detections and improves detection accuracy while optimizing power consumption.
Solution Approach 2:
The patent changes the detection parameters from static manually-set thresholds to dynamic ML-based decision boundaries. The system monitors multiple correlated attributes simultaneously and uses ML to determine optimal detection thresholds adaptively, improving detection reliability while reducing false positives that would otherwise increase power consumption and latency.
2Reliability
If conventional isolation techniques (regional routing algorithms) are used to isolate HT-infected components, then HT isolation capability is provided, but network throughput is limited and network latency increases
Solution Approach 1:
The patent implements dynamic isolation techniques where the routing algorithm adapts in real-time based on HT detection results and network conditions. Instead of static regional routing that permanently blocks certain channels, the system dynamically adjusts packet routes to avoid infected regions while maintaining optimal throughput, allowing flexible adaptation to changing threat landscapes.
Solution Approach 2:
The patent introduces an ML-based intermediary layer between the HT detection mechanism and the routing isolation mechanism. This intermediary analyzes detection results and network state to intelligently determine isolation strategies, mediating between security requirements and performance optimization to minimize throughput impact while maintaining effective HT isolation.
3Ease of operation
If manual threshold selection is used for HT detection, then detection simplicity is maintained, but detection accuracy decreases due to false detection and misdetection
Solution Approach 1:
The patent implements a self-training ML system that automatically learns optimal detection thresholds from normal NoC operation data. The system performs self-service by continuously training on collected attributes (buffer utilization, link utilization, temperature, power consumption) and adapting to changing network conditions, eliminating the need for manual threshold tuning while achieving high detection accuracy.
Solution Approach 2:
The patent performs preliminary training of the ML model during system initialization or offline phases, preparing the detection system before actual HT monitoring begins. This preliminary action allows the system to learn normal operation patterns and establish baseline thresholds in advance, improving detection accuracy without adding complexity to the real-time detection operation.
Data Source
AI summary
Systems and methods are disclosed for improving on-chip security, while minimizes the latency and cost of security techniques to improve system-level performance and power simultaneously. The framework uses machine learning algorithms, such as an artificial neural network (ANN), for runtime attack detection with higher accuracy. Further, a learning-based attack mitigation method using deep reinforcement learning is disclosed, where the method may be used to isolate the malicious components and to optimize network latency and energy-efficiency.