SAFE-IoT: Attesting Firmware in IoT Swarms using Volatile Memory and a Mixture of Experts

Advances in 5G mobile networks and artificial intelligence have led to rapid growth in the Internet of Things (IoT) as part of various smart initiatives. Embedded IoT microcontrollers are an easy target of firmware and network attacks, which become the root cause of various node-level and device-to-device (D2D) propagated anomalies. Thus, firmware integrity is essential to ensuring IoT security. Although several existing techniques require a legitimate copy of the device’s firmware, authentic firmware may not be available. In addition, the available literature also has limitations in terms of scalability, computational complexity, low availability, and the need for specialized hardware. This paper presents Swarm Attestation of Firmware in Embedded-IoT(SAFE-IoT) to solve these problems using a Mixture of Denoising Autoencoder Networks (MoDAE) framework and is the first to use Static Random Access Memory (SRAM) to attest firmware in IoT swarms. We present a volatile memory dataset for swarm attestation, which contains thirteen network scenarios that capture various D2D relationships. SAFE-IoT achieves a 99+% attestation rate on authorized firmware, a 100% detection rate on anomalous firmware, and a 95+% detection rate on D2D propagated anomalies. The proposed method has a latency of 10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">−4</sup> seconds per node. Lastly, we analyze robustness against perturbation of SRAM traces.