Security of Smart Dust: Robust Key Derivation for Single-Chip Systems

The Smart Dust vision seeks to enable large networks of millimeter-scale wireless sensor nodes that tightly integrate sensing, computation, communication, and power management into a single-chip device. Establishing a robust hardware root of trust for such devices remains challenging, particularly in single, low-cost chip manufacturing processes that lack embedded writable Non-Volatile Memory (NVM) for secure key storage. The thesis addresses the problem of deriving long-lived cryptographic keys on single-chip devices under the assumption of powerful physical attackers without relying on process-specific NVM.

We begin by analyzing the attack surface of highly integrated single-chip devices compared with conventional board-based devices. We show how architectural differences change the feasibility and impact of physical attacks and highlight a structural constraint: many single-chip devices do not provide writable NVM. Even when NVM is available, its persistent nature makes stored keys an attractive target for physical attacks. 

The thesis then investigates SRAM-based Physically Unclonable Functions (PUFs) as a candidate root of trust. By experimentally characterizing the start-up behavior of on-chip SRAM on a representative single-chip platform, we evaluate PUF properties and show that suitably chosen memory regions can provide stable and distinctive responses that are adequate for PUF-based keys. Building on this, the thesis develops the Threshold-based Majority Voting Scheme (TMVS), a lightweight stabilization scheme tailored to Smart Dust constraints, together with a two-stage extension (TS-TMVS). Both use simple majority voting decoders to eliminate noise and mitigate bias in SRAM PUF responses, while avoiding the substantial entropy loss of repetition codes and the implementation complexity of heavy error-correcting codes. They still rely on helper data stored in NVM to support key reconstruction.

Finally, the thesis proposes On-Demand Helper Data (ODHD), an NVM-free key derivation method that regenerates both helper data and secret keys reliably by incorporating a limited number of SRAM measurements into the encoding procedure. Full experimental characterization is ongoing work. Taken together, these contributions demonstrate that robust key derivation from SRAM PUFs is feasible on single-chip devices and provide a concrete path towards secure roots of trust for future Smart Dust systems.

Présentation en Anglais (slides en Anglais)
Talk in English (slides in English)