Spin-based chip to set new standards for IoT devices protection

• Physical Unclonable Functions offer a compelling lightweight cryptographic alternative, serving as a foundational root of trust for resource-limited devices.

The proliferation of the Internet of Things (IoT) has engendered a parallel surge in hardware security vulnerabilities.

Due to cost constraints, many IoT devices lack sophisticated encryption capabilities, leaving sensitive data susceptible to malicious actors. Physical Unclonable Functions (PUFs) offer a compelling lightweight cryptographic alternative, serving as a foundational root of trust for these resource-limited devices.

Traditional CMOS-based PUFs, however, suffer from limitations including suboptimal randomness, high power consumption, significant area overhead, and susceptibility to environmental variations.

In response to these challenges, researchers from Beihang University and Truth Memory Corporation have engineered a novel solution: a 1 Kbit spin-orbit torque magnetic random access memory (SOT-MRAM) chip incorporating a PUF, fabricated using an 180nm complementary metal oxide semiconductor (CMOS) process.

Performance metrics

The SOT-MRAM sr-PUF presents a robust and reconfigurable PUF architecture designed to withstand machine-learning attacks.

The operational principle of the SOT-MRAM sr-PUF involves initialising the memory array with a specific writing voltage, carefully calibrated to balance the probabilities of high- and low-resistance states to approximately 50 per cent.

Subsequently, a computing-in-memory (CIM) approach is employed to generate challenge-response pairs (CRPs). By comparing current summations across different column combinations, a single-bit response is derived.

Performance metrics for the SOT-MRAM sr-PUF are noteworthy. The device exhibits a CRP capacity of 109, coupled with a uniformity of 50.07 per cent, diffuseness of 50 per cent, and uniqueness of 49.89 per cent.

Critically, the bit error rate remains at zero per cent, even at an elevated temperature of 375K. These near-ideal values underscore the PUF’s exceptional randomness and reliability.

Furthermore, the reconfigurability of the SOT-MRAM sr-PUF is a crucial attribute. Through the application of varying write voltages, the PUF can dynamically refresh its CRPs.

The reconfigurability is quantified by a Hamming distance of 49.31 per cent and a correlation coefficient below 0.2 between different reconfiguration cycles, significantly hindering attackers attempting to extract output keys via side-channel analysis.

In the face of machine-learning attacks, the SOT-MRAM sr-PUF demonstrates considerable resilience. Testing with logistic regression, support vector machines, and multilayer perceptrons yielded prediction accuracies approaching the ideal 50 per cent mark, indicating effective resistance against such threats.