Monolithic Security Primitive Integration in Self-Rectifying Memristors for Extreme-Temperature Internet-of-Things

Abstract Reliable hardware security primitives that withstand harsh temperatures are essential for next-generation Internet of Things (IoT) deployed in industrial, automotive, and remote environments, where thermal stress undermines data integrity and device authentication. Conventional silicon-based true random number generators (TRNGs) and physical unclonable functions (PUFs) suffer from catastrophic performance degradation above 85 °C. Here, we address this critical gap with a unified, temperature-agnostic security primitive based on a Pt/HfO2-y/WO3-x/TiN self-rectifying memristor (SRM). The device exhibits exceptional performance across an unprecedented thermal range (-50 to 150 °C), featuring ultrahigh rectification (>106), nonlinearity (>105), and stable non-volatility with >10-year data retention. By harnessing intrinsic random telegraph noise (RTN) and stochastic conductance decay dynamics, we monolithically integrate TRNG and PUF functionalities on a single 3232 crossbar array. The TRNG demonstrates resilience against machine learning attacks, maintaining near-ideal prediction accuracy at both 150 °C and -50 °C. Simultaneously, the PUF achieves a record-low bit error rate of 1.859×10⁻2% across this 200 °C temperature swing, outperforming all existing silicon-based solutions by a margin greater than 85 °C. Validation against generative adversarial network attacks and tripartite authentication protocols confirms this selector-free, memristor-based approach as a pivotal advancement for secure, scalable hardware security in extreme-temperature IoT systems.