Abstract
Side-channel analysis (SCA), which involves analyzing physical emissions from devices to infer sensitive information, was traditionally regarded as a technique for enabling cryptographic attacks. In recent years, SCA has been increasingly applied to defensive purposes, with anomaly detection emerging as a prominent application. This survey systematizes the expanding field of Side-Channel-based Anomaly Detection (SCAD), which utilizes physical emissions such as electromagnetic radiation, power consumption patterns, acoustic signals, and thermal signatures to enable real-time system profiling, authentication, attestation, and malware detection. In contrast to software-based methods, SCAD provides non-invasive monitoring of devices, including embedded systems deployed in critical infrastructures. As a result, SCAD approaches enhance system robustness against sophisticated threats, including code injection, hardware Trojans, and micro-architectural attacks. This survey reviews over 50 SCAD methodologies across multiple domains and device types. The complete detection workflow of SCAD processes, encompassing signal acquisition, reprocessing, feature extraction, and model evaluation, is examined in detail, and for each phase, dominant methodologies are compared, including statistical and machine learning approaches such as shallow learning and deep learning, with an emphasis on their respective strengths and limitations. Publicly available datasets are also discussed, along with experimental practices. Lastly, common pitfalls and emerging challenges are identified, including issues related to signal synchronization, the impact of environmental noise, robustness against adversarial attacks, and the need for scalable and explainable detection models.