Chip security that stays strong from Arctic cold to engine-bay heat

A car engine block, an orbiting satellite, and an uninsulated factory sensor all share a hidden vulnerability: intense heat quietly scrambles the electronic locks that safeguard our connected world. Conventional silicon-based security primitives, such as True Random Number Generators (TRNGs) and Physical Unclonable Functions (PUFs), rely on microscopic physical quirks to create uncopyable device IDs and random encryption keys. However, their internal circuits suffer from severe performance degradation and rising bit error rates above 85°C, leaving critical infrastructure vulnerable to spoofing and data theft.

In International Journal of Extreme Manufacturing, Prof. Yishu Zhang in Zhejiang University and his co-workers have now created a compact security component made of stacked metal oxides that produces reliable hardware fingerprints and cryptographic keys across a broad temperature span from -50°C to 150°C.

The central roadblock to building dense arrays of memory-resistors, known as memristors, has long been the problem of rogue current leaks. When these devices are packed into intersecting wire grids, electrical currents tend to creep backward along neighboring pathways, scrambling stored data unless engineers attach dedicated external gatekeeper transistors to every single junction. These extra components clutter the silicon wafer, ballooning fabrication costs and limiting miniaturization.

Prof. Zhang's team addressed this bottleneck by stacking a 10-nanometer film of hafnium oxide atop a 40-nanometer film of tungsten oxide, sandwiched neatly between platinum and titanium nitride contacts.

This multi-layer sandwich functions like an automatic microscopic turnstile. The natural energy mismatch across its interfaces lets forward signals pass freely while strangling backward electrical leaks down to a mere trillionth of an ampere, establishing a built-in one-way gate without requiring a separate control transistor.

Inside the oxide core, missing oxygen atoms act as stepping stones that line up under a voltage pulse to form a temporary conductive pathway. When the voltage turns off, these pathways naturally dissolve.

Because manufacturing leaves every single device with a slightly different microscopic texture, each bridge collapses at its own distinct pace. This unique decay time acts as a permanent hardware fingerprint, while the microscopic electrical fluctuations running through the signal provide a steady stream of random data for generating encryption keys.

Across an extreme 200°C testing range, spanning temperatures colder than an Antarctic blizzard to well beyond the boiling point of water, the system maintained stable security metrics. Its hardware fingerprint kept an error rate of 0.01859% at peak heat and ran with zero errors at room temperature across 100,000 test cycles without needing software corrections and achieved zero bit errors at room temperature across 100,000 test cycles without needing software error correction — a record for memristor-based PUFs. At the same time, attacking machine-learning algorithms could not predict the output of the random number generator, achieving only a 50% accuracy rate equivalent to guessing on a coin toss.

Because the design works without external selector transistors, chip foundries can manufacture these arrays using standard sputtering equipment already present in existing production lines. For the public, this translates into sturdier security hardware for automotive networks, power grids, and remote monitoring stations exposed to harsh field environments. The next step toward commercial adoption involves scaling these prototype 32-by-32 grids into larger multi-kilobit arrays and integrating the necessary signal-amplification circuits directly onto complete and packaged silicon chips.

International Journal of Extreme Manufacturing (IJEM, IF: 25.1) is devoted to publishing articles of the highest quality and significance to pushing the limits of scales, precision, performance and environments in manufacturing.

• Maintained #1 in Engineering, Manufacturing for 4 consecutive years
• Average time to First Decision after Peer Review: 34 days
• Open Access Publishing 

Visit our webpage, like us on Facebook, and follow us on Twitter and LinkedIn.