Topology-aware Detection and Localization of Distributed Denial-of-Service Attacks in Network-on-Chips
Hansika Weerasena, Xiaoguo Jia, Prabhat Mishra · 2025
Read Topology-aware Detection and Localization of Distributed Denial-of-Service Attacks in Network-on-Chips by Hansika Weerasena, Xiaoguo Jia, Prabhat Mishra (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-14898.txt.
Introduction
Hansika Weerasena, Xiaoguo Jia, and Prabhat Mishra present a defense against denial-of-service attacks inside computer chips, in a paper published in two thousand twenty-five. Modern chips route messages between cores through a network-on-chip. That shared fabric is a tempting target.
In a distributed denial-of-service attack, several malicious components flood the network together. Their shifting traffic patterns dodge static detection rules. The authors model the on-chip network as a graph and apply graph neural networks to its traffic. Because the method is topology-aware, it can both detect an attack and localize where it strikes. The framework targets heterogeneous, high-performance platforms where a flooded interconnect can cripple the whole system.
Catalog record
| CREATOR | Hansika Weerasena, Xiaoguo Jia, Prabhat Mishra |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-14898
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Topology-aware Detection and Localization of Distributed Denial-of-Service Attacks in Network-on-Chips Hansika Weerasena, Xiaoguo Jia, Prabhat Mishra Abstract . Network-on-Chip (NoC) enables on-chip communication between diverse cores in modern System-on-Chip (SoC) designs. With its shared communication fabric, NoC has become a focal point for various security threats, especially in heterogeneous and high-performance computing platforms. Among these attacks, Distributed Denial of Service (DDoS) attacks occur when multiple malicious entities collaborate to overwhelm and disrupt access to critical system components, potentially causing severe performance degradation or complete disruption of services. These attacks are particularly challenging to detect due to their distributed nature and dynamic traffic patterns in NoC, which often evade static detection rules or simple profiling. This paper presents a framework to conduct topology-aware detection and localization of DDoS attacks using Graph Neural Networks (GNNs) by analyzing NoC traffic patterns. Specifically, by modeling the NoC as a graph, our method utilizes spatiotemporal traffic features to effectively identify and localize DDoS attacks. Unlike prior works that rely on handcrafted features or threshold-based detection, our GNN-based approach operates directly on raw inter-flit delay data, learning complex traffic dependencies without manual intervention. Experimental results demonstrate that our approach can detect and localize DDoS attacks with high accuracy (up to 99%) while maintaining consistent performance under diverse attack strategies. Furthermore, the proposed method exhibits strong robustness across varying numbers and placements of malicious IPs, different packet injection rates, application workloads, and architectural configurations, including both 2D mesh and 3D TSV-based NoCs. Our work provides a scalable, flexible, and architecture-agnostic defense mechanism, significantly improving the availability and trustworthiness of on-chip communication in future SoC designs. Journal: TECS Volume: 1 1 1 1 CCS: Networks Network on chip CCS: Networks Denial-of-service attacks CCS: Security and privacy Security protocols CCS: Computing methodologies Machine learning 1. Introduction Parallel workloads and specialized computing accelerators, such as neural network accelerators, have become pivotal in advancing computing capabilities, significantly shaping modern processor architecture designs. These workloads demand fast, energy-efficient communication across a growing number of cores, often under tight latency constraints. Heterogeneous Systems-on-Chips (SoCs) and Multi-Processor SoCs (MPSoCs) now integrate a large number of Intellectual Property (IP) cores in a single chip, reflecting a major shift towards more complex and powerful systems with diverse compute and memory hierarchies. For example, Intel’s Xeon® Scalable Processor ( 1 ) supports up to 128 cores, and Altra® multicore server processors feature up to 192 cores ( 2 ) . Recent advancements in heterogeneous integration and 3D chip stacking are expected to push this boundary even further ( 3 ) , enabling ultra-dense core integration on future SoCs. To support this growing communication demand, Network-on-Chip (NoC) has emerged as the de facto standard for inter-core communication, meeting the performance, scalability, and modularity requirements of densely packed chips. NoC ensures high throughput and low latency by offering a robust, scalable, and distributed communication framework within the chip. For instance, Intel employs the Skylake Mesh NoC ( 1 ) in their server-grade processors to efficiently manage communication across many cores. Figure 1 illustrates a 4\times 4 mesh NoC comprising 16 IP cores. Each node in the NoC contains a router, a Network Interface (NI), and an IP core. When a core needs to communicate with another, it injects packets through its NI, which are routed hop-by-hop through intermediate routers to reach the destination. For example, when a processing core ( P ) initiates a memory request after a cache miss, it forwards the request to one of the four corner memory controllers, as shown in Figure 1 , highlighting the fundamental role of NoC in memory access and inter-core communication. Due to cost and time-to-market constraints, SoC manufacturers often use third-party vendors and services from the global supply chain ( 4 ) . Typically, only a few IP cores are designed in-house, while the others are reusable IPs from third-party vendors. This long supply chain introduces the risk of malicious implants through various channels, such as untrustworthy CAD tools, rogue designers, or foundries. Furthermore, the growing complexity of SoC designs makes comprehensive security verification increasingly difficult. Microelectronics and Advanced Packaging Technologies (MAPT) roadmap ( 5 ) by Semiconductor Research Corporation (SRC) introduces Heterogeneous Integration as the key concept for cost- and power-efficient design in next-generation computing systems. Heterogeneous Integration increases security concerns by expanding the attack surface due to its complexity. The scalability of Heterogeneous Integration enables the integration of on-chip Machine Learning (ML) accelerators, which can be utilized for ML-based traffic monitoring to enhance security. While designing energy-efficient NoCs is a primary goal, securing them is equally important since exploiting an NoC could give attackers access to shared inter-core communications and the ability to compromise the security of the entire computing infrastructure. Figure 1 . 4x4 mesh NoC topology. Only corner nodes have shared Memory Controllers (MCs). Malicious IPs (MIPs) can target MCs and flood packets to launch DDoS attack. Security threats across different NoC technologies (electrical, wireless, optical, and hybrid) have been widely explored in the literature ( 4 ) . Denial-of-Service (DoS) attacks target critical resources, overwhelming communication and processing, ultimately disrupting access to system services. Distributed DoS (DDoS) attacks leverage multiple malicious nodes, making them more effective and stealthy. Such attacks are especially problematic in systems with shared memory controllers, caches, or accelerators, where a bottleneck at a single point can cascade into global slowdowns. Figure 1 illustrates a DDoS attack where two malicious IPs (MIPs) overwhelm a shared Memory Controller (MC) by flooding it with bogus memory requests. Previous work ( 6 , 7 ) on mitigating DoS attacks in NoCs have focused on statistical traffic modeling, but these methods rely on statistical bounds, which fail to adapt to changing application characteristics. While existing traditional machine learning (ML)-based mitigation methods are promising ( 8 , 9 ) , they require extensive manual feature engineering and retraining as applications evolve. Furthermore, their reliance on handcrafted thresholds or congestion metrics makes them less effective under dynamic, stealthy DDoS behaviors. Moreover, some of these approaches ( 8 , 6 ) are limited to detecting DoS attacks with a single adversary. They also do not leverage the topological structure of NoC, which encodes spatial dependencies vital for effective threat modeling. Recent advances in acceleration of deep neural networks (DNNs) and heterogeneous integration with chiplets enable on-chip DNN acceleration for real-time traffic monitoring. This presents an opportunity to combine hardware-accelerated inference with deep learning models for adaptive and scalable runtime security enforcement in future NoC-based SoCs. In this paper, we utilize graph neural networks (GNNs) for scalable and topology-aware detection and localization of DDoS attacks in Network-on-Chip systems. Specifically, this paper makes the following major contributions: • We propose a novel methodology to detect and localize DDoS attacks in NoC architectures using GNNs, leveraging raw inter-flit delays and spatial graph structure to extract spatiotemporal features without manual tuning. • The proposed GNN-based detection and localization framework is rigorously evaluated across a wide range of traffic configurations, 2D and 3D architectures, attack scenarios, and benchmark applications, demonstrating its generalization and scalability This paper is organized as follows. Section 2 provides background on core concepts and surveys related efforts. Section 3 outlines the threat model. Section 4 describes our proposed framework