MacLeR: Machine Learning-based Run-Time Hardware Trojan Detection in Resource-Constrained IoT Edge Devices

Faiq Khalid, Syed Rafay Hasan, Sara Zia, Osman Hasan, Falah Awwad, Muhammad Shafique

Abstract

Traditional learning-based approaches for run-time Hardware Trojan detection require complex and expensive on-chip data acquisition frameworks and thus incur high area and power overhead. To address these challenges, we propose to leverage the power correlation between the executing instructions of a microprocessor to establish a machine learning-based run-time Hardware Trojan (HT) detection framework, called MacLeR. To reduce the overhead of data acquisition, we propose a single power-port current acquisition block using current sensors in time-division multiplexing, which increases accuracy while incurring reduced area overhead. Our experimental results show that compared to state-of-the-art HT detection techniques, MacLeR achieves 10% better HT detection accuracy (i.e., 96.256%) while incurring a 7x reduction in area and power overhead (i.e., 0.025% of the area of the SoC and <0.07% of the power of the SoC). Furthermore, MacLeR demonstrates significant robustness against process variation and aging, with an accuracy drop ~10x less than competing ML-based techniques.

Introduction

As the semiconductor supply chain becomes increasingly globalized, the integration of third-party Intellectual Property (IP) blocks has become standard practice. However, this outsourcing introduces the risk of Hardware Trojans (HTs)—malicious circuit modifications that can leak sensitive data or cause catastrophic system failure. Detecting these Trojans at runtime is particularly challenging for resource-constrained IoT edge devices, where strict power and area budgets make traditional monitoring hardware impractical. Additionally, manufacturing inconsistencies (process variation) and circuit aging can create "noise" that mimics Trojan behavior, leading to false positives or missed detections.

The MacLeR Framework

MacLeR addresses these challenges through a lightweight, high-resolution monitoring architecture. The framework is built upon three technical pillars:

  1. Fine-Grained Power Profiling: Instead of monitoring aggregate chip power, MacLeR analyzes the power consumption of individual pipeline stages (Fetch, Decode, Execute, etc.) for specific instruction groups. This provides a high-resolution "fingerprint" of normal operation.
  2. Single Power-Port Current Acquisition Block (SP-CAB): To minimize hardware overhead, MacLeR utilizes a novel hardware block featuring current mirrors and Time-Division Multiplexing (TDM). This allows a single monitoring port to capture data from multiple pipeline stages sequentially, drastically reducing silicon area requirements.
  3. Lightweight Machine Learning: The system employs a simple Multi-Layer Perceptron (MLP) with only two hidden layers. To optimize training, instructions are categorized into five functional groups based on the SPARC V8 architecture, ensuring the model remains efficient enough for edge deployment.
The MacLeR runtime monitoring architecture.

Fine-Grained Power Analysis and Trojan Impact

A core insight of MacLeR is the Interaction Hypothesis: malicious activity in an untrusted IP will have a detectable impact on the power behavior of a trusted IP it interacts with. By monitoring a known-good processor core, MacLeR can identify Trojans hidden within "black box" third-party components.

Trojans often trigger during specific instruction phases. For instance, a Trojan might activate during the decode stage of a MOV instruction or the execute stage of an ADD instruction. By monitoring at the pipeline level, MacLeR captures these transient power "spikes" that would otherwise be lost in the average total power consumption of the chip.

Power profiles and deviations across different instruction stages and Trojan scenarios.

Experimental Evaluation

The researchers evaluated MacLeR using a LEON3-based SoC integrated with various IPs (AES, RSA, Ethernet, VGA) and standard HT benchmarks from Trust-Hub.

  • Detection Accuracy: MacLeR achieved a peak accuracy of 96.256%, outperforming existing ML-based side-channel techniques by approximately 10%.
  • Resource Efficiency: The monitoring hardware occupies a mere 0.025% of the total SoC area and consumes less than 0.07% of total power, representing a 7x reduction in overhead compared to prior state-of-the-art methods.
  • Environmental Robustness: Under worst-case 10-year aging and process variation models (TSMC 65nm), MacLeR's accuracy dropped by less than 9%—a 10x improvement in stability over competing ML models.

Conclusion

MacLeR provides a highly efficient and robust solution for runtime Hardware Trojan detection in IoT devices. While future work is needed to address complexities like pipeline stalls and ultra-high-speed sampling requirements, the current framework demonstrates that fine-grained power profiling combined with time-multiplexed sensing can effectively secure resource-constrained hardware against sophisticated supply chain attacks.

Summary updated

Abstract

Traditional learning-based approaches for run-time Hardware Trojan detection require complex and expensive on-chip data acquisition frameworks and thus incur high area and power overhead. To address these challenges, we propose to leverage the power correlation between the executing instructions of a microprocessor to establish a machine learning-based run-time Hardware Trojan (HT) detection framework, called MacLeR. To reduce the overhead of data acquisition, we propose a single power-port current acquisition block using current sensors in time-division multiplexing, which increases accuracy while incurring reduced area overhead. We have implemented a practical solution by analyzing multiple HT benchmarks inserted in the RTL of a system-on-chip (SoC) consisting of four LEON3 processors integrated with other IPs like vga_lcd, RSA, AES, Ethernet, and memory controllers. Our experimental results show that compared to state-of-the-art HT detection techniques, MacLeR achieves 10\% better HT detection accuracy (i.e., 96.256%) while incurring a 7x reduction in area and power overhead (i.e., 0.025% of the area of the SoC and <0.07% of the power of the SoC). In addition, we also analyze the impact of process variation and aging on the extracted power profiles and the HT detection accuracy of MacLeR. Our analysis shows that variations in fine-grained power profiles due to the HTs are significantly higher compared to the variations in fine-grained power profiles caused by the process variations (PV) and aging effects. Moreover, our analysis demonstrates that, on average, the HT detection accuracy drop in MacLeR is less than 1% and 9% when considering only PV and PV with worst-case aging, respectively, which is ~10x less than in the case of the state-of-the-art ML-based HT detection technique.
BibTeX
@article{khalid2020macler,
  author = {Faiq Khalid and Syed Rafay Hasan and Sara Zia and Osman Hasan and Falah Awwad and Muhammad Shafique},
  title = {{MacLeR: Machine Learning-based Run-Time Hardware Trojan Detection in Resource-Constrained IoT Edge Devices}},
  journal = {IEEE Trans. Comput. Aided Des. Integr. Circuits Syst.},
  year = {2020},
  doi = {10.1109/tcad.2020.3012236},
}
Figure