Reference-Free Spectral Analysis of EM Side-Channels for Always-on Hardware Trojan Detection

TL;DR

This work addresses the challenge of detecting always-on hardware Trojans without golden references by introducing a reference-free framework that analyzes EM side-channels through multi-resolution time–frequency representations. It combines passive EM data collection with $STFT$ across multiple window sizes and unsupervised $GMM$ modeling, using $BIC$ for model-order selection, and detects Trojans via cross-scale consistency of mixture structures. A key finding is that HT-free designs display scale-dependent variability in the number of mixture components, while always-on HTs produce persistent, low-complexity spectral footprints largely invariant across scales. The approach enables non-invasive, reference-free HT detection suitable for post-deployment and supply-chain-constrained environments, with demonstration on an $AES$-128 workload.

Abstract

Always-on hardware Trojans (HTs) pose a critical risk to trusted microelectronics, yet most side-channel detection methods rely on unavailable golden references. We present a reference-free approach that combines time-frequency EM analysis with Gaussian Mixture Models (GMMs). By applying Short-Time Fourier Transform (STFT) at multiple window sizes, we show that HT-free circuits exhibit fluctuating statistical structure, while always-on HTs leave persistent footprints with fewer, more consistent mixture components. Results on AES-128 demonstrate feasibility without requiring reference models.

Figures (7)

• Figure 1: Overview of the proposed reference-free hardware Trojan detection framework, organized into four conceptual steps. (1) Passive EM data acquisition captures side-channel emissions without golden references. (2) Time–frequency representations are generated using multi-resolution STFT to expose transient and persistent spectral behavior. (3) Unsupervised Gaussian Mixture Models (GMMs) with BIC-based model order selection capture statistical structure at each scale. (4) Detection is performed by evaluating cross-scale consistency, where always-on Trojans exhibit fewer and more stable mixture components than Trojan-free designs.
• Figure 2: Experimental setup.
• Figure 3: Spectrogram analysis of HT-free AES-128 after applying 500 fixed key and plaintext with the segment length of 200 for STFT.
• Figure 4: Spectrogram analysis of HT-inserted AES-128 after applying 500 fixed key and plaintext with the segment length of 200 for STFT.
• Figure 5: Feature vectors constructed from frequency components and their corresponding stability-map magnitudes for the HT-free AES-128 implementation across multiple STFT window sizes. Each subfigure shows the distribution of stability scores over time and frequency for a given window length.