EM Fingerprints: Towards Identifying Unauthorized Hardware Substitutions in the Supply Chain Jungle

Abstract

This paper proposes a system capable of branding digital device components based on the EM signals typically emitted during their normal operational cycles. Such signals contain digital artifacts that are unique, which may act as an identifier of a particular device component e.g., its CPU, or the entire device if one chooses to take into account a combination of multiple such components. In real-life scenarios, this "biometrical" fingerprinting of hardware has to be conducted only once, possibly as part of an initial device configuration process with minimum additional maintenance time and cost, by the network administrators. At a subsequent stage, devices can get "authenticated" by comparing their newly emitted signals against the preexisting database during routine checks. The experimental results attest that the proposed approach can effectively protect a network against unrecognized potentially rogue devices posing as benign or malicious substitutions of hardware components at the chip level with near-perfect accuracy. One may view the proposed system as a technical solution to verify the trustworthiness of digital parts as well as the actors involved in certain stages of the supply chain.144 2020 Symposium on Security and Privacy Workshops (SPW)

Towards Scalable Anomaly Detection for Embedded Devices through Synthetic EM Fingerprinting

“…Moreover, depending on the type of antenna, the approach can be less invasive as the monitoring can be performed from a distance in real time. In fact, EM-based anomaly detection tools have proven to be successful for the detection of extensive [7], [8], or even minimal modifications, say, down to the injection of a few instructions (at the assembly level) [9], [10].…”

On the Limits of EM Based Detection of Control Logic Injection Attacks In Noisy Environments

“…In the past, alternative sidechannels have been considered, including the power consumption patterns [3], [4], [5], the thermal footprint [6] or the acoustic signals [7] of devices during their operation. Nevertheless, EM-based approaches [8], [9], [10], [11], [12] offer a comparative advantage since the signals themselves can be captured and analyzed in a completely non-intrusive fashion, i.e., no installation of software in the monitored device is assumed. Moreover, unlike analog signals describing the analysis of power consumption, the EM spectrum offers high bandwidth.…”